@inproceedings{13053, abstract = {Deep neural networks (DNNs) often have to be compressed, via pruning and/or quantization, before they can be deployed in practical settings. In this work we propose a new compression-aware minimizer dubbed CrAM that modifies the optimization step in a principled way, in order to produce models whose local loss behavior is stable under compression operations such as pruning. Thus, dense models trained via CrAM should be compressible post-training, in a single step, without significant accuracy loss. Experimental results on standard benchmarks, such as residual networks for ImageNet classification and BERT models for language modelling, show that CrAM produces dense models that can be more accurate than the standard SGD/Adam-based baselines, but which are stable under weight pruning: specifically, we can prune models in one-shot to 70-80% sparsity with almost no accuracy loss, and to 90% with reasonable (∼1%) accuracy loss, which is competitive with gradual compression methods. Additionally, CrAM can produce sparse models which perform well for transfer learning, and it also works for semi-structured 2:4 pruning patterns supported by GPU hardware. The code for reproducing the results is available at this https URL .}, author = {Peste, Elena-Alexandra and Vladu, Adrian and Kurtic, Eldar and Lampert, Christoph and Alistarh, Dan-Adrian}, booktitle = {11th International Conference on Learning Representations }, location = {Kigali, Rwanda }, title = {{CrAM: A Compression-Aware Minimizer}}, year = {2023}, } @phdthesis{13074, abstract = {Deep learning has become an integral part of a large number of important applications, and many of the recent breakthroughs have been enabled by the ability to train very large models, capable to capture complex patterns and relationships from the data. At the same time, the massive sizes of modern deep learning models have made their deployment to smaller devices more challenging; this is particularly important, as in many applications the users rely on accurate deep learning predictions, but they only have access to devices with limited memory and compute power. One solution to this problem is to prune neural networks, by setting as many of their parameters as possible to zero, to obtain accurate sparse models with lower memory footprint. Despite the great research progress in obtaining sparse models that preserve accuracy, while satisfying memory and computational constraints, there are still many challenges associated with efficiently training sparse models, as well as understanding their generalization properties. The focus of this thesis is to investigate how the training process of sparse models can be made more efficient, and to understand the differences between sparse and dense models in terms of how well they can generalize to changes in the data distribution. We first study a method for co-training sparse and dense models, at a lower cost compared to regular training. With our method we can obtain very accurate sparse networks, and dense models that can recover the baseline accuracy. Furthermore, we are able to more easily analyze the differences, at prediction level, between the sparse-dense model pairs. Next, we investigate the generalization properties of sparse neural networks in more detail, by studying how well different sparse models trained on a larger task can adapt to smaller, more specialized tasks, in a transfer learning scenario. Our analysis across multiple pruning methods and sparsity levels reveals that sparse models provide features that can transfer similarly to or better than the dense baseline. However, the choice of the pruning method plays an important role, and can influence the results when the features are fixed (linear finetuning), or when they are allowed to adapt to the new task (full finetuning). Using sparse models with fixed masks for finetuning on new tasks has an important practical advantage, as it enables training neural networks on smaller devices. However, one drawback of current pruning methods is that the entire training cycle has to be repeated to obtain the initial sparse model, for every sparsity target; in consequence, the entire training process is costly and also multiple models need to be stored. In the last part of the thesis we propose a method that can train accurate dense models that are compressible in a single step, to multiple sparsity levels, without additional finetuning. Our method results in sparse models that can be competitive with existing pruning methods, and which can also successfully generalize to new tasks.}, author = {Peste, Elena-Alexandra}, issn = {2663-337X}, pages = {147}, publisher = {Institute of Science and Technology Austria}, title = {{Efficiency and generalization of sparse neural networks}}, doi = {10.15479/at:ista:13074}, year = {2023}, } @article{14320, abstract = {The development of two-dimensional materials has resulted in a diverse range of novel, high-quality compounds with increasing complexity. A key requirement for a comprehensive quantitative theory is the accurate determination of these materials' band structure parameters. However, this task is challenging due to the intricate band structures and the indirect nature of experimental probes. In this work, we introduce a general framework to derive band structure parameters from experimental data using deep neural networks. We applied our method to the penetration field capacitance measurement of trilayer graphene, an effective probe of its density of states. First, we demonstrate that a trained deep network gives accurate predictions for the penetration field capacitance as a function of tight-binding parameters. Next, we use the fast and accurate predictions from the trained network to automatically determine tight-binding parameters directly from experimental data, with extracted parameters being in a good agreement with values in the literature. We conclude by discussing potential applications of our method to other materials and experimental techniques beyond penetration field capacitance.}, author = {Henderson, Paul M and Ghazaryan, Areg and Zibrov, Alexander A. and Young, Andrea F. and Serbyn, Maksym}, issn = {2469-9969}, journal = {Physical Review B}, number = {12}, publisher = {American Physical Society}, title = {{Deep learning extraction of band structure parameters from density of states: A case study on trilayer graphene}}, doi = {10.1103/physrevb.108.125411}, volume = {108}, year = {2023}, } @inproceedings{14410, abstract = {This paper focuses on the implementation details of the baseline methods and a recent lightweight conditional model extrapolation algorithm LIMES [5] for streaming data under class-prior shift. LIMES achieves superior performance over the baseline methods, especially concerning the minimum-across-day accuracy, which is important for the users of the system. In this work, the key measures to facilitate reproducibility and enhance the credibility of the results are described.}, author = {Tomaszewska, Paulina and Lampert, Christoph}, booktitle = {International Workshop on Reproducible Research in Pattern Recognition}, isbn = {9783031407727}, issn = {1611-3349}, location = {Montreal, Canada}, pages = {67--73}, publisher = {Springer Nature}, title = {{On the implementation of baselines and lightweight conditional model extrapolation (LIMES) under class-prior shift}}, doi = {10.1007/978-3-031-40773-4_6}, volume = {14068}, year = {2023}, } @article{14446, abstract = {Recent work has paid close attention to the first principle of Granger causality, according to which cause precedes effect. In this context, the question may arise whether the detected direction of causality also reverses after the time reversal of unidirectionally coupled data. Recently, it has been shown that for unidirectionally causally connected autoregressive (AR) processes X → Y, after time reversal of data, the opposite causal direction Y → X is indeed detected, although typically as part of the bidirectional X↔ Y link. As we argue here, the answer is different when the measured data are not from AR processes but from linked deterministic systems. When the goal is the usual forward data analysis, cross-mapping-like approaches correctly detect X → Y, while Granger causality-like approaches, which should not be used for deterministic time series, detect causal independence X → Y. The results of backward causal analysis depend on the predictability of the reversed data. Unlike AR processes, observables from deterministic dynamical systems, even complex nonlinear ones, can be predicted well forward, while backward predictions can be difficult (notably when the time reversal of a function leads to one-to-many relations). To address this problem, we propose an approach based on models that provide multiple candidate predictions for the target, combined with a loss function that consideres only the best candidate. The resulting good forward and backward predictability supports the view that unidirectionally causally linked deterministic dynamical systems X → Y can be expected to detect the same link both before and after time reversal.}, author = {Jakubík, Jozef and Bui Thi Mai, Phuong and Chvosteková, Martina and Krakovská, Anna}, issn = {1335-8871}, journal = {Measurement Science Review}, number = {4}, pages = {175--183}, publisher = {Sciendo}, title = {{Against the flow of time with multi-output models}}, doi = {10.2478/msr-2023-0023}, volume = {23}, year = {2023}, } @inproceedings{14771, abstract = {Pruning—that is, setting a significant subset of the parameters of a neural network to zero—is one of the most popular methods of model compression. Yet, several recent works have raised the issue that pruning may induce or exacerbate bias in the output of the compressed model. Despite existing evidence for this phenomenon, the relationship between neural network pruning and induced bias is not well-understood. In this work, we systematically investigate and characterize this phenomenon in Convolutional Neural Networks for computer vision. First, we show that it is in fact possible to obtain highly-sparse models, e.g. with less than 10% remaining weights, which do not decrease in accuracy nor substantially increase in bias when compared to dense models. At the same time, we also find that, at higher sparsities, pruned models exhibit higher uncertainty in their outputs, as well as increased correlations, which we directly link to increased bias. We propose easy-to-use criteria which, based only on the uncompressed model, establish whether bias will increase with pruning, and identify the samples most susceptible to biased predictions post-compression. Our code can be found at https://github.com/IST-DASLab/pruned-vision-model-bias.}, author = {Iofinova, Eugenia B and Peste, Elena-Alexandra and Alistarh, Dan-Adrian}, booktitle = {2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition}, issn = {2575-7075}, location = {Vancouver, BC, Canada}, pages = {24364--24373}, publisher = {IEEE}, title = {{Bias in pruned vision models: In-depth analysis and countermeasures}}, doi = {10.1109/cvpr52729.2023.02334}, year = {2023}, } @inproceedings{14921, abstract = {Neural collapse (NC) refers to the surprising structure of the last layer of deep neural networks in the terminal phase of gradient descent training. Recently, an increasing amount of experimental evidence has pointed to the propagation of NC to earlier layers of neural networks. However, while the NC in the last layer is well studied theoretically, much less is known about its multi-layered counterpart - deep neural collapse (DNC). In particular, existing work focuses either on linear layers or only on the last two layers at the price of an extra assumption. Our paper fills this gap by generalizing the established analytical framework for NC - the unconstrained features model - to multiple non-linear layers. Our key technical contribution is to show that, in a deep unconstrained features model, the unique global optimum for binary classification exhibits all the properties typical of DNC. This explains the existing experimental evidence of DNC. We also empirically show that (i) by optimizing deep unconstrained features models via gradient descent, the resulting solution agrees well with our theory, and (ii) trained networks recover the unconstrained features suitable for the occurrence of DNC, thus supporting the validity of this modeling principle.}, author = {Súkeník, Peter and Mondelli, Marco and Lampert, Christoph}, booktitle = {37th Annual Conference on Neural Information Processing Systems}, location = {New Orleans, LA, United States}, title = {{Deep neural collapse is provably optimal for the deep unconstrained features model}}, year = {2023}, } @unpublished{15039, abstract = {A crucial property for achieving secure, trustworthy and interpretable deep learning systems is their robustness: small changes to a system's inputs should not result in large changes to its outputs. Mathematically, this means one strives for networks with a small Lipschitz constant. Several recent works have focused on how to construct such Lipschitz networks, typically by imposing constraints on the weight matrices. In this work, we study an orthogonal aspect, namely the role of the activation function. We show that commonly used activation functions, such as MaxMin, as well as all piece-wise linear ones with two segments unnecessarily restrict the class of representable functions, even in the simplest one-dimensional setting. We furthermore introduce the new N-activation function that is provably more expressive than currently popular activation functions. We provide code at this https URL.}, author = {Prach, Bernd and Lampert, Christoph}, booktitle = {arXiv}, title = {{1-Lipschitz neural networks are more expressive with N-activations}}, doi = {10.48550/ARXIV.2311.06103}, year = {2023}, } @unpublished{12660, abstract = {We present Cross-Client Label Propagation(XCLP), a new method for transductive federated learning. XCLP estimates a data graph jointly from the data of multiple clients and computes labels for the unlabeled data by propagating label information across the graph. To avoid clients having to share their data with anyone, XCLP employs two cryptographically secure protocols: secure Hamming distance computation and secure summation. We demonstrate two distinct applications of XCLP within federated learning. In the first, we use it in a one-shot way to predict labels for unseen test points. In the second, we use it to repeatedly pseudo-label unlabeled training data in a federated semi-supervised setting. Experiments on both real federated and standard benchmark datasets show that in both applications XCLP achieves higher classification accuracy than alternative approaches.}, author = {Scott, Jonathan A and Yeo, Michelle X and Lampert, Christoph}, booktitle = {arXiv}, title = {{Cross-client Label Propagation for transductive federated learning}}, doi = {10.48550/arXiv.2210.06434}, year = {2022}, } @unpublished{12662, abstract = {Modern machine learning tasks often require considering not just one but multiple objectives. For example, besides the prediction quality, this could be the efficiency, robustness or fairness of the learned models, or any of their combinations. Multi-objective learning offers a natural framework for handling such problems without having to commit to early trade-offs. Surprisingly, statistical learning theory so far offers almost no insight into the generalization properties of multi-objective learning. In this work, we make first steps to fill this gap: we establish foundational generalization bounds for the multi-objective setting as well as generalization and excess bounds for learning with scalarizations. We also provide the first theoretical analysis of the relation between the Pareto-optimal sets of the true objectives and the Pareto-optimal sets of their empirical approximations from training data. In particular, we show a surprising asymmetry: all Pareto-optimal solutions can be approximated by empirically Pareto-optimal ones, but not vice versa.}, author = {Súkeník, Peter and Lampert, Christoph}, booktitle = {arXiv}, title = {{Generalization in Multi-objective machine learning}}, doi = {10.48550/arXiv.2208.13499}, year = {2022}, } @article{12495, abstract = {Fairness-aware learning aims at constructing classifiers that not only make accurate predictions, but also do not discriminate against specific groups. It is a fast-growing area of machine learning with far-reaching societal impact. However, existing fair learning methods are vulnerable to accidental or malicious artifacts in the training data, which can cause them to unknowingly produce unfair classifiers. In this work we address the problem of fair learning from unreliable training data in the robust multisource setting, where the available training data comes from multiple sources, a fraction of which might not be representative of the true data distribution. We introduce FLEA, a filtering-based algorithm that identifies and suppresses those data sources that would have a negative impact on fairness or accuracy if they were used for training. As such, FLEA is not a replacement of prior fairness-aware learning methods but rather an augmentation that makes any of them robust against unreliable training data. We show the effectiveness of our approach by a diverse range of experiments on multiple datasets. Additionally, we prove formally that –given enough data– FLEA protects the learner against corruptions as long as the fraction of affected data sources is less than half. Our source code and documentation are available at https://github.com/ISTAustria-CVML/FLEA.}, author = {Iofinova, Eugenia B and Konstantinov, Nikola H and Lampert, Christoph}, issn = {2835-8856}, journal = {Transactions on Machine Learning Research}, publisher = {ML Research Press}, title = {{FLEA: Provably robust fair multisource learning from unreliable training data}}, year = {2022}, } @inproceedings{11839, abstract = {It is a highly desirable property for deep networks to be robust against small input changes. One popular way to achieve this property is by designing networks with a small Lipschitz constant. In this work, we propose a new technique for constructing such Lipschitz networks that has a number of desirable properties: it can be applied to any linear network layer (fully-connected or convolutional), it provides formal guarantees on the Lipschitz constant, it is easy to implement and efficient to run, and it can be combined with any training objective and optimization method. In fact, our technique is the first one in the literature that achieves all of these properties simultaneously. Our main contribution is a rescaling-based weight matrix parametrization that guarantees each network layer to have a Lipschitz constant of at most 1 and results in the learned weight matrices to be close to orthogonal. Hence we call such layers almost-orthogonal Lipschitz (AOL). Experiments and ablation studies in the context of image classification with certified robust accuracy confirm that AOL layers achieve results that are on par with most existing methods. Yet, they are simpler to implement and more broadly applicable, because they do not require computationally expensive matrix orthogonalization or inversion steps as part of the network architecture. We provide code at https://github.com/berndprach/AOL.}, author = {Prach, Bernd and Lampert, Christoph}, booktitle = {Computer Vision – ECCV 2022}, isbn = {9783031198021}, location = {Tel Aviv, Israel}, pages = {350--365}, publisher = {Springer Nature}, title = {{Almost-orthogonal layers for efficient general-purpose Lipschitz networks}}, doi = {10.1007/978-3-031-19803-8_21}, volume = {13681}, year = {2022}, } @inproceedings{10752, abstract = {The digitalization of almost all aspects of our everyday lives has led to unprecedented amounts of data being freely available on the Internet. In particular social media platforms provide rich sources of user-generated data, though typically in unstructured form, and with high diversity, such as written in many different languages. Automatically identifying meaningful information in such big data resources and extracting it efficiently is one of the ongoing challenges of our time. A common step for this is sentiment analysis, which forms the foundation for tasks such as opinion mining or trend prediction. Unfortunately, publicly available tools for this task are almost exclusively available for English-language texts. Consequently, a large fraction of the Internet users, who do not communicate in English, are ignored in automatized studies, a phenomenon called rare-language discrimination.In this work we propose a technique to overcome this problem by a truly multi-lingual model, which can be trained automatically without linguistic knowledge or even the ability to read the many target languages. The main step is to combine self-annotation, specifically the use of emoticons as a proxy for labels, with multi-lingual sentence representations.To evaluate our method we curated several large datasets from data obtained via the free Twitter streaming API. The results show that our proposed multi-lingual training is able to achieve sentiment predictions at the same quality level for rare languages as for frequent ones, and in particular clearly better than what mono-lingual training achieves on the same data. }, author = {Lampert, Jasmin and Lampert, Christoph}, booktitle = {2021 IEEE International Conference on Big Data}, isbn = {9781665439022}, location = {Orlando, FL, United States}, pages = {5185--5192}, publisher = {IEEE}, title = {{Overcoming rare-language discrimination in multi-lingual sentiment analysis}}, doi = {10.1109/bigdata52589.2021.9672003}, year = {2022}, } @inproceedings{12161, abstract = {We introduce LIMES, a new method for learning with non-stationary streaming data, inspired by the recent success of meta-learning. The main idea is not to attempt to learn a single classifier that would have to work well across all occurring data distributions, nor many separate classifiers, but to exploit a hybrid strategy: we learn a single set of model parameters from which a specific classifier for any specific data distribution is derived via classifier adaptation. Assuming a multiclass classification setting with class-prior shift, the adaptation step can be performed analytically with only the classifier’s bias terms being affected. Another contribution of our work is an extrapolation step that predicts suitable adaptation parameters for future time steps based on the previous data. In combination, we obtain a lightweight procedure for learning from streaming data with varying class distribution that adds no trainable parameters and almost no memory or computational overhead compared to training a single model. Experiments on a set of exemplary tasks using Twitter data show that LIMES achieves higher accuracy than alternative approaches, especially with respect to the relevant real-world metric of lowest within-day accuracy.}, author = {Tomaszewska, Paulina and Lampert, Christoph}, booktitle = {26th International Conference on Pattern Recognition}, issn = {2831-7475}, location = {Montreal, Canada}, pages = {2128--2134}, publisher = {Institute of Electrical and Electronics Engineers}, title = {{Lightweight conditional model extrapolation for streaming data under class-prior shift}}, doi = {10.1109/icpr56361.2022.9956195}, volume = {2022}, year = {2022}, } @inproceedings{12299, abstract = {Transfer learning is a classic paradigm by which models pretrained on large “upstream” datasets are adapted to yield good results on “downstream” specialized datasets. Generally, more accurate models on the “upstream” dataset tend to provide better transfer accuracy “downstream”. In this work, we perform an in-depth investigation of this phenomenon in the context of convolutional neural networks (CNNs) trained on the ImageNet dataset, which have been pruned-that is, compressed by sparsifiying their connections. We consider transfer using unstructured pruned models obtained by applying several state-of-the-art pruning methods, including magnitude-based, second-order, regrowth, lottery-ticket, and regularization approaches, in the context of twelve standard transfer tasks. In a nutshell, our study shows that sparse models can match or even outperform the transfer performance of dense models, even at high sparsities, and, while doing so, can lead to significant inference and even training speedups. At the same time, we observe and analyze significant differences in the behaviour of different pruning methods. The code is available at: https://github.com/IST-DASLab/sparse-imagenet-transfer.}, author = {Iofinova, Eugenia B and Peste, Elena-Alexandra and Kurtz, Mark and Alistarh, Dan-Adrian}, booktitle = {2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition}, issn = {2575-7075}, location = {New Orleans, LA, United States}, pages = {12256--12266}, publisher = {Institute of Electrical and Electronics Engineers}, title = {{How well do sparse ImageNet models transfer?}}, doi = {10.1109/cvpr52688.2022.01195}, year = {2022}, } @article{10802, abstract = {Addressing fairness concerns about machine learning models is a crucial step towards their long-term adoption in real-world automated systems. While many approaches have been developed for training fair models from data, little is known about the robustness of these methods to data corruption. In this work we consider fairness-aware learning under worst-case data manipulations. We show that an adversary can in some situations force any learner to return an overly biased classifier, regardless of the sample size and with or without degrading accuracy, and that the strength of the excess bias increases for learning problems with underrepresented protected groups in the data. We also prove that our hardness results are tight up to constant factors. To this end, we study two natural learning algorithms that optimize for both accuracy and fairness and show that these algorithms enjoy guarantees that are order-optimal in terms of the corruption ratio and the protected groups frequencies in the large data limit.}, author = {Konstantinov, Nikola H and Lampert, Christoph}, issn = {1533-7928}, journal = {Journal of Machine Learning Research}, keywords = {Fairness, robustness, data poisoning, trustworthy machine learning, PAC learning}, pages = {1--60}, publisher = {ML Research Press}, title = {{Fairness-aware PAC learning from corrupted data}}, volume = {23}, year = {2022}, } @inproceedings{13241, abstract = {Addressing fairness concerns about machine learning models is a crucial step towards their long-term adoption in real-world automated systems. Many approaches for training fair models from data have been developed and an implicit assumption about such algorithms is that they are able to recover a fair model, despite potential historical biases in the data. In this work we show a number of impossibility results that indicate that there is no learning algorithm that can recover a fair model when a proportion of the dataset is subject to arbitrary manipulations. Specifically, we prove that there are situations in which an adversary can force any learner to return a biased classifier, with or without degrading accuracy, and that the strength of this bias increases for learning problems with underrepresented protected groups in the data. Our results emphasize on the importance of studying further data corruption models of various strength and of establishing stricter data collection practices for fairness-aware learning.}, author = {Konstantinov, Nikola H and Lampert, Christoph}, booktitle = {Proceedings of Machine Learning Research}, issn = {2640-3498}, pages = {59--83}, publisher = {ML Research Press}, title = {{On the impossibility of fairness-aware learning from corrupted data}}, volume = {171}, year = {2022}, } @phdthesis{10799, abstract = {Because of the increasing popularity of machine learning methods, it is becoming important to understand the impact of learned components on automated decision-making systems and to guarantee that their consequences are beneficial to society. In other words, it is necessary to ensure that machine learning is sufficiently trustworthy to be used in real-world applications. This thesis studies two properties of machine learning models that are highly desirable for the sake of reliability: robustness and fairness. In the first part of the thesis we study the robustness of learning algorithms to training data corruption. Previous work has shown that machine learning models are vulnerable to a range of training set issues, varying from label noise through systematic biases to worst-case data manipulations. This is an especially relevant problem from a present perspective, since modern machine learning methods are particularly data hungry and therefore practitioners often have to rely on data collected from various external sources, e.g. from the Internet, from app users or via crowdsourcing. Naturally, such sources vary greatly in the quality and reliability of the data they provide. With these considerations in mind, we study the problem of designing machine learning algorithms that are robust to corruptions in data coming from multiple sources. We show that, in contrast to the case of a single dataset with outliers, successful learning within this model is possible both theoretically and practically, even under worst-case data corruptions. The second part of this thesis deals with fairness-aware machine learning. There are multiple areas where machine learning models have shown promising results, but where careful considerations are required, in order to avoid discrimanative decisions taken by such learned components. Ensuring fairness can be particularly challenging, because real-world training datasets are expected to contain various forms of historical bias that may affect the learning process. In this thesis we show that data corruption can indeed render the problem of achieving fairness impossible, by tightly characterizing the theoretical limits of fair learning under worst-case data manipulations. However, assuming access to clean data, we also show how fairness-aware learning can be made practical in contexts beyond binary classification, in particular in the challenging learning to rank setting.}, author = {Konstantinov, Nikola H}, isbn = {978-3-99078-015-2}, issn = {2663-337X}, keywords = {robustness, fairness, machine learning, PAC learning, adversarial learning}, pages = {176}, publisher = {Institute of Science and Technology Austria}, title = {{Robustness and fairness in machine learning}}, doi = {10.15479/at:ista:10799}, year = {2022}, } @inproceedings{9210, abstract = {Modern neural networks can easily fit their training set perfectly. Surprisingly, despite being “overfit” in this way, they tend to generalize well to future data, thereby defying the classic bias–variance trade-off of machine learning theory. Of the many possible explanations, a prevalent one is that training by stochastic gradient descent (SGD) imposes an implicit bias that leads it to learn simple functions, and these simple functions generalize well. However, the specifics of this implicit bias are not well understood. In this work, we explore the smoothness conjecture which states that SGD is implicitly biased towards learning functions that are smooth. We propose several measures to formalize the intuitive notion of smoothness, and we conduct experiments to determine whether SGD indeed implicitly optimizes for these measures. Our findings rule out the possibility that smoothness measures based on first-order derivatives are being implicitly enforced. They are supportive, though, of the smoothness conjecture for measures based on second-order derivatives.}, author = {Volhejn, Vaclav and Lampert, Christoph}, booktitle = {42nd German Conference on Pattern Recognition}, isbn = {9783030712778}, issn = {1611-3349}, location = {Tübingen, Germany}, pages = {246--259}, publisher = {Springer}, title = {{Does SGD implicitly optimize for smoothness?}}, doi = {10.1007/978-3-030-71278-5_18}, volume = {12544}, year = {2021}, } @inproceedings{9416, abstract = {We study the inductive bias of two-layer ReLU networks trained by gradient flow. We identify a class of easy-to-learn (`orthogonally separable') datasets, and characterise the solution that ReLU networks trained on such datasets converge to. Irrespective of network width, the solution turns out to be a combination of two max-margin classifiers: one corresponding to the positive data subset and one corresponding to the negative data subset. The proof is based on the recently introduced concept of extremal sectors, for which we prove a number of properties in the context of orthogonal separability. In particular, we prove stationarity of activation patterns from some time onwards, which enables a reduction of the ReLU network to an ensemble of linear subnetworks.}, author = {Bui Thi Mai, Phuong and Lampert, Christoph}, booktitle = {9th International Conference on Learning Representations}, location = {Virtual}, title = {{The inductive bias of ReLU networks on orthogonally separable data}}, year = {2021}, } @unpublished{10803, abstract = {Given the abundance of applications of ranking in recent years, addressing fairness concerns around automated ranking systems becomes necessary for increasing the trust among end-users. Previous work on fair ranking has mostly focused on application-specific fairness notions, often tailored to online advertising, and it rarely considers learning as part of the process. In this work, we show how to transfer numerous fairness notions from binary classification to a learning to rank setting. Our formalism allows us to design methods for incorporating fairness objectives with provable generalization guarantees. An extensive experimental evaluation shows that our method can improve ranking fairness substantially with no or only little loss of model quality.}, author = {Konstantinov, Nikola H and Lampert, Christoph}, booktitle = {arXiv}, title = {{Fairness through regularization for learning to rank}}, doi = {10.48550/arXiv.2102.05996}, year = {2021}, } @phdthesis{9418, abstract = {Deep learning is best known for its empirical success across a wide range of applications spanning computer vision, natural language processing and speech. Of equal significance, though perhaps less known, are its ramifications for learning theory: deep networks have been observed to perform surprisingly well in the high-capacity regime, aka the overfitting or underspecified regime. Classically, this regime on the far right of the bias-variance curve is associated with poor generalisation; however, recent experiments with deep networks challenge this view. This thesis is devoted to investigating various aspects of underspecification in deep learning. First, we argue that deep learning models are underspecified on two levels: a) any given training dataset can be fit by many different functions, and b) any given function can be expressed by many different parameter configurations. We refer to the second kind of underspecification as parameterisation redundancy and we precisely characterise its extent. Second, we characterise the implicit criteria (the inductive bias) that guide learning in the underspecified regime. Specifically, we consider a nonlinear but tractable classification setting, and show that given the choice, neural networks learn classifiers with a large margin. Third, we consider learning scenarios where the inductive bias is not by itself sufficient to deal with underspecification. We then study different ways of ‘tightening the specification’: i) In the setting of representation learning with variational autoencoders, we propose a hand- crafted regulariser based on mutual information. ii) In the setting of binary classification, we consider soft-label (real-valued) supervision. We derive a generalisation bound for linear networks supervised in this way and verify that soft labels facilitate fast learning. Finally, we explore an application of soft-label supervision to the training of multi-exit models.}, author = {Bui Thi Mai, Phuong}, issn = {2663-337X}, pages = {125}, publisher = {Institute of Science and Technology Austria}, title = {{Underspecification in deep learning}}, doi = {10.15479/AT:ISTA:9418}, year = {2021}, } @inbook{14987, abstract = {The goal of zero-shot learning is to construct a classifier that can identify object classes for which no training examples are available. When training data for some of the object classes is available but not for others, the name generalized zero-shot learning is commonly used. In a wider sense, the phrase zero-shot is also used to describe other machine learning-based approaches that require no training data from the problem of interest, such as zero-shot action recognition or zero-shot machine translation.}, author = {Lampert, Christoph}, booktitle = {Computer Vision}, editor = {Ikeuchi, Katsushi}, isbn = {9783030634155}, pages = {1395--1397}, publisher = {Springer}, title = {{Zero-Shot Learning}}, doi = {10.1007/978-3-030-63416-2_874}, year = {2021}, } @unpublished{8063, abstract = {We present a generative model of images that explicitly reasons over the set of objects they show. Our model learns a structured latent representation that separates objects from each other and from the background; unlike prior works, it explicitly represents the 2D position and depth of each object, as well as an embedding of its segmentation mask and appearance. The model can be trained from images alone in a purely unsupervised fashion without the need for object masks or depth information. Moreover, it always generates complete objects, even though a significant fraction of training images contain occlusions. Finally, we show that our model can infer decompositions of novel images into their constituent objects, including accurate prediction of depth ordering and segmentation of occluded parts.}, author = {Anciukevicius, Titas and Lampert, Christoph and Henderson, Paul M}, booktitle = {arXiv}, title = {{Object-centric image generation with factored depths, locations, and appearances}}, year = {2020}, } @inproceedings{8188, abstract = {A natural approach to generative modeling of videos is to represent them as a composition of moving objects. Recent works model a set of 2D sprites over a slowly-varying background, but without considering the underlying 3D scene that gives rise to them. We instead propose to model a video as the view seen while moving through a scene with multiple 3D objects and a 3D background. Our model is trained from monocular videos without any supervision, yet learns to generate coherent 3D scenes containing several moving objects. We conduct detailed experiments on two datasets, going beyond the visual complexity supported by state-of-the-art generative approaches. We evaluate our method on depth-prediction and 3D object detection---tasks which cannot be addressed by those earlier works---and show it out-performs them even on 2D instance segmentation and tracking.}, author = {Henderson, Paul M and Lampert, Christoph}, booktitle = {34th Conference on Neural Information Processing Systems}, isbn = {9781713829546}, location = {Vancouver, Canada}, pages = {3106–3117}, publisher = {Curran Associates}, title = {{Unsupervised object-centric video generation and decomposition in 3D}}, volume = {33}, year = {2020}, } @article{6952, abstract = {We present a unified framework tackling two problems: class-specific 3D reconstruction from a single image, and generation of new 3D shape samples. These tasks have received considerable attention recently; however, most existing approaches rely on 3D supervision, annotation of 2D images with keypoints or poses, and/or training with multiple views of each object instance. Our framework is very general: it can be trained in similar settings to existing approaches, while also supporting weaker supervision. Importantly, it can be trained purely from 2D images, without pose annotations, and with only a single view per instance. We employ meshes as an output representation, instead of voxels used in most prior work. This allows us to reason over lighting parameters and exploit shading information during training, which previous 2D-supervised methods cannot. Thus, our method can learn to generate and reconstruct concave object classes. We evaluate our approach in various settings, showing that: (i) it learns to disentangle shape from pose and lighting; (ii) using shading in the loss improves performance compared to just silhouettes; (iii) when using a standard single white light, our model outperforms state-of-the-art 2D-supervised methods, both with and without pose supervision, thanks to exploiting shading cues; (iv) performance improves further when using multiple coloured lights, even approaching that of state-of-the-art 3D-supervised methods; (v) shapes produced by our model capture smooth surfaces and fine details better than voxel-based approaches; and (vi) our approach supports concave classes such as bathtubs and sofas, which methods based on silhouettes cannot learn.}, author = {Henderson, Paul M and Ferrari, Vittorio}, issn = {1573-1405}, journal = {International Journal of Computer Vision}, pages = {835--854}, publisher = {Springer Nature}, title = {{Learning single-image 3D reconstruction by generative modelling of shape, pose and shading}}, doi = {10.1007/s11263-019-01219-8}, volume = {128}, year = {2020}, } @inproceedings{7936, abstract = {State-of-the-art detection systems are generally evaluated on their ability to exhaustively retrieve objects densely distributed in the image, across a wide variety of appearances and semantic categories. Orthogonal to this, many real-life object detection applications, for example in remote sensing, instead require dealing with large images that contain only a few small objects of a single class, scattered heterogeneously across the space. In addition, they are often subject to strict computational constraints, such as limited battery capacity and computing power.To tackle these more practical scenarios, we propose a novel flexible detection scheme that efficiently adapts to variable object sizes and densities: We rely on a sequence of detection stages, each of which has the ability to predict groups of objects as well as individuals. Similar to a detection cascade, this multi-stage architecture spares computational effort by discarding large irrelevant regions of the image early during the detection process. The ability to group objects provides further computational and memory savings, as it allows working with lower image resolutions in early stages, where groups are more easily detected than individuals, as they are more salient. We report experimental results on two aerial image datasets, and show that the proposed method is as accurate yet computationally more efficient than standard single-shot detectors, consistently across three different backbone architectures.}, author = {Royer, Amélie and Lampert, Christoph}, booktitle = {IEEE Winter Conference on Applications of Computer Vision}, isbn = {9781728165530}, location = { Snowmass Village, CO, United States}, publisher = {IEEE}, title = {{Localizing grouped instances for efficient detection in low-resource scenarios}}, doi = {10.1109/WACV45572.2020.9093288}, year = {2020}, } @inproceedings{7937, abstract = {Fine-tuning is a popular way of exploiting knowledge contained in a pre-trained convolutional network for a new visual recognition task. However, the orthogonal setting of transferring knowledge from a pretrained network to a visually different yet semantically close source is rarely considered: This commonly happens with real-life data, which is not necessarily as clean as the training source (noise, geometric transformations, different modalities, etc.).To tackle such scenarios, we introduce a new, generalized form of fine-tuning, called flex-tuning, in which any individual unit (e.g. layer) of a network can be tuned, and the most promising one is chosen automatically. In order to make the method appealing for practical use, we propose two lightweight and faster selection procedures that prove to be good approximations in practice. We study these selection criteria empirically across a variety of domain shifts and data scarcity scenarios, and show that fine-tuning individual units, despite its simplicity, yields very good results as an adaptation technique. As it turns out, in contrast to common practice, rather than the last fully-connected unit it is best to tune an intermediate or early one in many domain- shift scenarios, which is accurately detected by flex-tuning.}, author = {Royer, Amélie and Lampert, Christoph}, booktitle = {2020 IEEE Winter Conference on Applications of Computer Vision}, isbn = {9781728165530}, location = {Snowmass Village, CO, United States}, publisher = {IEEE}, title = {{A flexible selection scheme for minimum-effort transfer learning}}, doi = {10.1109/WACV45572.2020.9093635}, year = {2020}, } @inbook{8092, abstract = {Image translation refers to the task of mapping images from a visual domain to another. Given two unpaired collections of images, we aim to learn a mapping between the corpus-level style of each collection, while preserving semantic content shared across the two domains. We introduce xgan, a dual adversarial auto-encoder, which captures a shared representation of the common domain semantic content in an unsupervised way, while jointly learning the domain-to-domain image translations in both directions. We exploit ideas from the domain adaptation literature and define a semantic consistency loss which encourages the learned embedding to preserve semantics shared across domains. We report promising qualitative results for the task of face-to-cartoon translation. The cartoon dataset we collected for this purpose, “CartoonSet”, is also publicly available as a new benchmark for semantic style transfer at https://google.github.io/cartoonset/index.html.}, author = {Royer, Amélie and Bousmalis, Konstantinos and Gouws, Stephan and Bertsch, Fred and Mosseri, Inbar and Cole, Forrester and Murphy, Kevin}, booktitle = {Domain Adaptation for Visual Understanding}, editor = {Singh, Richa and Vatsa, Mayank and Patel, Vishal M. and Ratha, Nalini}, isbn = {9783030306717}, pages = {33--49}, publisher = {Springer Nature}, title = {{XGAN: Unsupervised image-to-image translation for many-to-many mappings}}, doi = {10.1007/978-3-030-30671-7_3}, year = {2020}, } @inproceedings{7481, abstract = {We address the following question: How redundant is the parameterisation of ReLU networks? Specifically, we consider transformations of the weight space which leave the function implemented by the network intact. Two such transformations are known for feed-forward architectures: permutation of neurons within a layer, and positive scaling of all incoming weights of a neuron coupled with inverse scaling of its outgoing weights. In this work, we show for architectures with non-increasing widths that permutation and scaling are in fact the only function-preserving weight transformations. For any eligible architecture we give an explicit construction of a neural network such that any other network that implements the same function can be obtained from the original one by the application of permutations and rescaling. The proof relies on a geometric understanding of boundaries between linear regions of ReLU networks, and we hope the developed mathematical tools are of independent interest.}, author = {Bui Thi Mai, Phuong and Lampert, Christoph}, booktitle = {8th International Conference on Learning Representations}, location = {Online}, title = {{Functional vs. parametric equivalence of ReLU networks}}, year = {2020}, } @inproceedings{8724, abstract = {We study the problem of learning from multiple untrusted data sources, a scenario of increasing practical relevance given the recent emergence of crowdsourcing and collaborative learning paradigms. Specifically, we analyze the situation in which a learning system obtains datasets from multiple sources, some of which might be biased or even adversarially perturbed. It is known that in the single-source case, an adversary with the power to corrupt a fixed fraction of the training data can prevent PAC-learnability, that is, even in the limit of infinitely much training data, no learning system can approach the optimal test error. In this work we show that, surprisingly, the same is not true in the multi-source setting, where the adversary can arbitrarily corrupt a fixed fraction of the data sources. Our main results are a generalization bound that provides finite-sample guarantees for this learning setting, as well as corresponding lower bounds. Besides establishing PAC-learnability our results also show that in a cooperative learning setting sharing data with other parties has provable benefits, even if some participants are malicious. }, author = {Konstantinov, Nikola H and Frantar, Elias and Alistarh, Dan-Adrian and Lampert, Christoph}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, issn = {2640-3498}, location = {Online}, pages = {5416--5425}, publisher = {ML Research Press}, title = {{On the sample complexity of adversarial multi-source PAC learning}}, volume = {119}, year = {2020}, } @phdthesis{8390, abstract = {Deep neural networks have established a new standard for data-dependent feature extraction pipelines in the Computer Vision literature. Despite their remarkable performance in the standard supervised learning scenario, i.e. when models are trained with labeled data and tested on samples that follow a similar distribution, neural networks have been shown to struggle with more advanced generalization abilities, such as transferring knowledge across visually different domains, or generalizing to new unseen combinations of known concepts. In this thesis we argue that, in contrast to the usual black-box behavior of neural networks, leveraging more structured internal representations is a promising direction for tackling such problems. In particular, we focus on two forms of structure. First, we tackle modularity: We show that (i) compositional architectures are a natural tool for modeling reasoning tasks, in that they efficiently capture their combinatorial nature, which is key for generalizing beyond the compositions seen during training. We investigate how to to learn such models, both formally and experimentally, for the task of abstract visual reasoning. Then, we show that (ii) in some settings, modularity allows us to efficiently break down complex tasks into smaller, easier, modules, thereby improving computational efficiency; We study this behavior in the context of generative models for colorization, as well as for small objects detection. Secondly, we investigate the inherently layered structure of representations learned by neural networks, and analyze its role in the context of transfer learning and domain adaptation across visually dissimilar domains. }, author = {Royer, Amélie}, isbn = {978-3-99078-007-7}, issn = {2663-337X}, pages = {197}, publisher = {Institute of Science and Technology Austria}, title = {{Leveraging structure in Computer Vision tasks for flexible Deep Learning models}}, doi = {10.15479/AT:ISTA:8390}, year = {2020}, } @inproceedings{8186, abstract = {Numerous methods have been proposed for probabilistic generative modelling of 3D objects. However, none of these is able to produce textured objects, which renders them of limited use for practical tasks. In this work, we present the first generative model of textured 3D meshes. Training such a model would traditionally require a large dataset of textured meshes, but unfortunately, existing datasets of meshes lack detailed textures. We instead propose a new training methodology that allows learning from collections of 2D images without any 3D information. To do so, we train our model to explain a distribution of images by modelling each image as a 3D foreground object placed in front of a 2D background. Thus, it learns to generate meshes that when rendered, produce images similar to those in its training set. A well-known problem when generating meshes with deep networks is the emergence of self-intersections, which are problematic for many use-cases. As a second contribution we therefore introduce a new generation process for 3D meshes that guarantees no self-intersections arise, based on the physical intuition that faces should push one another out of the way as they move. We conduct extensive experiments on our approach, reporting quantitative and qualitative results on both synthetic data and natural images. These show our method successfully learns to generate plausible and diverse textured 3D samples for five challenging object classes.}, author = {Henderson, Paul M and Tsiminaki, Vagia and Lampert, Christoph}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, issn = {2575-7075}, location = {Virtual}, pages = {7498--7507}, publisher = {IEEE}, title = {{Leveraging 2D data to learn textured 3D mesh generation}}, doi = {10.1109/CVPR42600.2020.00752}, year = {2020}, } @article{6944, abstract = {We study the problem of automatically detecting if a given multi-class classifier operates outside of its specifications (out-of-specs), i.e. on input data from a different distribution than what it was trained for. This is an important problem to solve on the road towards creating reliable computer vision systems for real-world applications, because the quality of a classifier’s predictions cannot be guaranteed if it operates out-of-specs. Previously proposed methods for out-of-specs detection make decisions on the level of single inputs. This, however, is insufficient to achieve low false positive rate and high false negative rates at the same time. In this work, we describe a new procedure named KS(conf), based on statistical reasoning. Its main component is a classical Kolmogorov–Smirnov test that is applied to the set of predicted confidence values for batches of samples. Working with batches instead of single samples allows increasing the true positive rate without negatively affecting the false positive rate, thereby overcoming a crucial limitation of single sample tests. We show by extensive experiments using a variety of convolutional network architectures and datasets that KS(conf) reliably detects out-of-specs situations even under conditions where other tests fail. It furthermore has a number of properties that make it an excellent candidate for practical deployment: it is easy to implement, adds almost no overhead to the system, works with any classifier that outputs confidence scores, and requires no a priori knowledge about how the data distribution could change.}, author = {Sun, Rémy and Lampert, Christoph}, issn = {1573-1405}, journal = {International Journal of Computer Vision}, number = {4}, pages = {970--995}, publisher = {Springer Nature}, title = {{KS(conf): A light-weight test if a multiclass classifier operates outside of its specifications}}, doi = {10.1007/s11263-019-01232-x}, volume = {128}, year = {2020}, } @book{7171, abstract = {Wissen Sie, was sich hinter künstlicher Intelligenz und maschinellem Lernen verbirgt? Dieses Sachbuch erklärt Ihnen leicht verständlich und ohne komplizierte Formeln die grundlegenden Methoden und Vorgehensweisen des maschinellen Lernens. Mathematisches Vorwissen ist dafür nicht nötig. Kurzweilig und informativ illustriert Lisa, die Protagonistin des Buches, diese anhand von Alltagssituationen. Ein Buch für alle, die in Diskussionen über Chancen und Risiken der aktuellen Entwicklung der künstlichen Intelligenz und des maschinellen Lernens mit Faktenwissen punkten möchten. Auch für Schülerinnen und Schüler geeignet!}, editor = {Kersting, Kristian and Lampert, Christoph and Rothkopf, Constantin}, isbn = {978-3-658-26762-9}, pages = {XIV, 245}, publisher = {Springer Nature}, title = {{Wie Maschinen Lernen: Künstliche Intelligenz Verständlich Erklärt}}, doi = {10.1007/978-3-658-26763-6}, year = {2019}, } @inproceedings{6942, abstract = {Graph games and Markov decision processes (MDPs) are standard models in reactive synthesis and verification of probabilistic systems with nondeterminism. The class of 𝜔 -regular winning conditions; e.g., safety, reachability, liveness, parity conditions; provides a robust and expressive specification formalism for properties that arise in analysis of reactive systems. The resolutions of nondeterminism in games and MDPs are represented as strategies, and we consider succinct representation of such strategies. The decision-tree data structure from machine learning retains the flavor of decisions of strategies and allows entropy-based minimization to obtain succinct trees. However, in contrast to traditional machine-learning problems where small errors are allowed, for winning strategies in graph games and MDPs no error is allowed, and the decision tree must represent the entire strategy. In this work we propose decision trees with linear classifiers for representation of strategies in graph games and MDPs. We have implemented strategy representation using this data structure and we present experimental results for problems on graph games and MDPs, which show that this new data structure presents a much more efficient strategy representation as compared to standard decision trees.}, author = {Ashok, Pranav and Brázdil, Tomáš and Chatterjee, Krishnendu and Křetínský, Jan and Lampert, Christoph and Toman, Viktor}, booktitle = {16th International Conference on Quantitative Evaluation of Systems}, isbn = {9783030302801}, issn = {0302-9743}, location = {Glasgow, United Kingdom}, pages = {109--128}, publisher = {Springer Nature}, title = {{Strategy representation by decision trees with linear classifiers}}, doi = {10.1007/978-3-030-30281-8_7}, volume = {11785}, year = {2019}, } @article{6554, abstract = {Due to the importance of zero-shot learning, i.e. classifying images where there is a lack of labeled training data, the number of proposed approaches has recently increased steadily. We argue that it is time to take a step back and to analyze the status quo of the area. The purpose of this paper is three-fold. First, given the fact that there is no agreed upon zero-shot learning benchmark, we first define a new benchmark by unifying both the evaluation protocols and data splits of publicly available datasets used for this task. This is an important contribution as published results are often not comparable and sometimes even flawed due to, e.g. pre-training on zero-shot test classes. Moreover, we propose a new zero-shot learning dataset, the Animals with Attributes 2 (AWA2) dataset which we make publicly available both in terms of image features and the images themselves. Second, we compare and analyze a significant number of the state-of-the-art methods in depth, both in the classic zero-shot setting but also in the more realistic generalized zero-shot setting. Finally, we discuss in detail the limitations of the current status of the area which can be taken as a basis for advancing it.}, author = {Xian, Yongqin and Lampert, Christoph and Schiele, Bernt and Akata, Zeynep}, issn = {1939-3539}, journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence}, number = {9}, pages = {2251 -- 2265}, publisher = {Institute of Electrical and Electronics Engineers (IEEE)}, title = {{Zero-shot learning - A comprehensive evaluation of the good, the bad and the ugly}}, doi = {10.1109/tpami.2018.2857768}, volume = {41}, year = {2019}, } @inproceedings{7479, abstract = {Multi-exit architectures, in which a stack of processing layers is interleaved with early output layers, allow the processing of a test example to stop early and thus save computation time and/or energy. In this work, we propose a new training procedure for multi-exit architectures based on the principle of knowledge distillation. The method encourage searly exits to mimic later, more accurate exits, by matching their output probabilities. Experiments on CIFAR100 and ImageNet show that distillation-based training significantly improves the accuracy of early exits while maintaining state-of-the-art accuracy for late ones. The method is particularly beneficial when training data is limited and it allows a straightforward extension to semi-supervised learning,i.e. making use of unlabeled data at training time. Moreover, it takes only afew lines to implement and incurs almost no computational overhead at training time, and none at all at test time.}, author = {Bui Thi Mai, Phuong and Lampert, Christoph}, booktitle = {IEEE International Conference on Computer Vision}, isbn = {9781728148038}, issn = {15505499}, location = {Seoul, Korea}, pages = {1355--1364}, publisher = {IEEE}, title = {{Distillation-based training for multi-exit architectures}}, doi = {10.1109/ICCV.2019.00144}, volume = {2019-October}, year = {2019}, } @inproceedings{7640, abstract = {We propose a new model for detecting visual relationships, such as "person riding motorcycle" or "bottle on table". This task is an important step towards comprehensive structured mage understanding, going beyond detecting individual objects. Our main novelty is a Box Attention mechanism that allows to model pairwise interactions between objects using standard object detection pipelines. The resulting model is conceptually clean, expressive and relies on well-justified training and prediction procedures. Moreover, unlike previously proposed approaches, our model does not introduce any additional complex components or hyperparameters on top of those already required by the underlying detection model. We conduct an experimental evaluation on two datasets, V-COCO and Open Images, demonstrating strong quantitative and qualitative results.}, author = {Kolesnikov, Alexander and Kuznetsova, Alina and Lampert, Christoph and Ferrari, Vittorio}, booktitle = {Proceedings of the 2019 International Conference on Computer Vision Workshop}, isbn = {9781728150239}, location = {Seoul, South Korea}, publisher = {IEEE}, title = {{Detecting visual relationships using box attention}}, doi = {10.1109/ICCVW.2019.00217}, year = {2019}, } @inproceedings{6569, abstract = {Knowledge distillation, i.e. one classifier being trained on the outputs of another classifier, is an empirically very successful technique for knowledge transfer between classifiers. It has even been observed that classifiers learn much faster and more reliably if trained with the outputs of another classifier as soft labels, instead of from ground truth data. So far, however, there is no satisfactory theoretical explanation of this phenomenon. In this work, we provide the first insights into the working mechanisms of distillation by studying the special case of linear and deep linear classifiers. Specifically, we prove a generalization bound that establishes fast convergence of the expected risk of a distillation-trained linear classifier. From the bound and its proof we extract three keyfactors that determine the success of distillation: data geometry – geometric properties of the datadistribution, in particular class separation, has an immediate influence on the convergence speed of the risk; optimization bias– gradient descentoptimization finds a very favorable minimum of the distillation objective; and strong monotonicity– the expected risk of the student classifier always decreases when the size of the training set grows.}, author = {Bui Thi Mai, Phuong and Lampert, Christoph}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, location = {Long Beach, CA, United States}, pages = {5142--5151}, publisher = {ML Research Press}, title = {{Towards understanding knowledge distillation}}, volume = {97}, year = {2019}, } @inproceedings{6590, abstract = {Modern machine learning methods often require more data for training than a single expert can provide. Therefore, it has become a standard procedure to collect data from external sources, e.g. via crowdsourcing. Unfortunately, the quality of these sources is not always guaranteed. As additional complications, the data might be stored in a distributed way, or might even have to remain private. In this work, we address the question of how to learn robustly in such scenarios. Studying the problem through the lens of statistical learning theory, we derive a procedure that allows for learning from all available sources, yet automatically suppresses irrelevant or corrupted data. We show by extensive experiments that our method provides significant improvements over alternative approaches from robust statistics and distributed optimization. }, author = {Konstantinov, Nikola H and Lampert, Christoph}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, location = {Long Beach, CA, USA}, pages = {3488--3498}, publisher = {ML Research Press}, title = {{Robust learning from untrusted sources}}, volume = {97}, year = {2019}, } @inproceedings{6482, abstract = {Computer vision systems for automatic image categorization have become accurate and reliable enough that they can run continuously for days or even years as components of real-world commercial applications. A major open problem in this context, however, is quality control. Good classification performance can only be expected if systems run under the specific conditions, in particular data distributions, that they were trained for. Surprisingly, none of the currently used deep network architectures have a built-in functionality that could detect if a network operates on data from a distribution it was not trained for, such that potentially a warning to the human users could be triggered. In this work, we describe KS(conf), a procedure for detecting such outside of specifications (out-of-specs) operation, based on statistical testing of the network outputs. We show by extensive experiments using the ImageNet, AwA2 and DAVIS datasets on a variety of ConvNets architectures that KS(conf) reliably detects out-of-specs situations. It furthermore has a number of properties that make it a promising candidate for practical deployment: it is easy to implement, adds almost no overhead to the system, works with all networks, including pretrained ones, and requires no a priori knowledge of how the data distribution could change. }, author = {Sun, Rémy and Lampert, Christoph}, isbn = {9783030129385}, issn = {1611-3349}, location = {Stuttgart, Germany}, pages = {244--259}, publisher = {Springer Nature}, title = {{KS(conf): A light-weight test if a ConvNet operates outside of Its specifications}}, doi = {10.1007/978-3-030-12939-2_18}, volume = {11269}, year = {2019}, } @phdthesis{68, abstract = {The most common assumption made in statistical learning theory is the assumption of the independent and identically distributed (i.i.d.) data. While being very convenient mathematically, it is often very clearly violated in practice. This disparity between the machine learning theory and applications underlies a growing demand in the development of algorithms that learn from dependent data and theory that can provide generalization guarantees similar to the independent situations. This thesis is dedicated to two variants of dependencies that can arise in practice. One is a dependence on the level of samples in a single learning task. Another dependency type arises in the multi-task setting when the tasks are dependent on each other even though the data for them can be i.i.d. In both cases we model the data (samples or tasks) as stochastic processes and introduce new algorithms for both settings that take into account and exploit the resulting dependencies. We prove the theoretical guarantees on the performance of the introduced algorithms under different evaluation criteria and, in addition, we compliment the theoretical study by the empirical one, where we evaluate some of the algorithms on two real world datasets to highlight their practical applicability.}, author = {Zimin, Alexander}, issn = {2663-337X}, pages = {92}, publisher = {Institute of Science and Technology Austria}, title = {{Learning from dependent data}}, doi = {10.15479/AT:ISTA:TH1048}, year = {2018}, } @phdthesis{197, abstract = {Modern computer vision systems heavily rely on statistical machine learning models, which typically require large amounts of labeled data to be learned reliably. Moreover, very recently computer vision research widely adopted techniques for representation learning, which further increase the demand for labeled data. However, for many important practical problems there is relatively small amount of labeled data available, so it is problematic to leverage full potential of the representation learning methods. One way to overcome this obstacle is to invest substantial resources into producing large labelled datasets. Unfortunately, this can be prohibitively expensive in practice. In this thesis we focus on the alternative way of tackling the aforementioned issue. We concentrate on methods, which make use of weakly-labeled or even unlabeled data. Specifically, the first half of the thesis is dedicated to the semantic image segmentation task. We develop a technique, which achieves competitive segmentation performance and only requires annotations in a form of global image-level labels instead of dense segmentation masks. Subsequently, we present a new methodology, which further improves segmentation performance by leveraging tiny additional feedback from a human annotator. By using our methods practitioners can greatly reduce the amount of data annotation effort, which is required to learn modern image segmentation models. In the second half of the thesis we focus on methods for learning from unlabeled visual data. We study a family of autoregressive models for modeling structure of natural images and discuss potential applications of these models. Moreover, we conduct in-depth study of one of these applications, where we develop the state-of-the-art model for the probabilistic image colorization task.}, author = {Kolesnikov, Alexander}, issn = {2663-337X}, pages = {113}, publisher = {Institute of Science and Technology Austria}, title = {{Weakly-Supervised Segmentation and Unsupervised Modeling of Natural Images}}, doi = {10.15479/AT:ISTA:th_1021}, year = {2018}, } @article{563, abstract = {In continuous populations with local migration, nearby pairs of individuals have on average more similar genotypes than geographically well separated pairs. A barrier to gene flow distorts this classical pattern of isolation by distance. Genetic similarity is decreased for sample pairs on different sides of the barrier and increased for pairs on the same side near the barrier. Here, we introduce an inference scheme that utilizes this signal to detect and estimate the strength of a linear barrier to gene flow in two-dimensions. We use a diffusion approximation to model the effects of a barrier on the geographical spread of ancestry backwards in time. This approach allows us to calculate the chance of recent coalescence and probability of identity by descent. We introduce an inference scheme that fits these theoretical results to the geographical covariance structure of bialleleic genetic markers. It can estimate the strength of the barrier as well as several demographic parameters. We investigate the power of our inference scheme to detect barriers by applying it to a wide range of simulated data. We also showcase an example application to a Antirrhinum majus (snapdragon) flower color hybrid zone, where we do not detect any signal of a strong genome wide barrier to gene flow.}, author = {Ringbauer, Harald and Kolesnikov, Alexander and Field, David and Barton, Nicholas H}, journal = {Genetics}, number = {3}, pages = {1231--1245}, publisher = {Genetics Society of America}, title = {{Estimating barriers to gene flow from distorted isolation-by-distance patterns}}, doi = {10.1534/genetics.117.300638}, volume = {208}, year = {2018}, } @article{321, abstract = {The twelve papers in this special section focus on learning systems with shared information for computer vision and multimedia communication analysis. In the real world, a realistic setting for computer vision or multimedia recognition problems is that we have some classes containing lots of training data and many classes containing a small amount of training data. Therefore, how to use frequent classes to help learning rare classes for which it is harder to collect the training data is an open question. Learning with shared information is an emerging topic in machine learning, computer vision and multimedia analysis. There are different levels of components that can be shared during concept modeling and machine learning stages, such as sharing generic object parts, sharing attributes, sharing transformations, sharing regularization parameters and sharing training examples, etc. Regarding the specific methods, multi-task learning, transfer learning and deep learning can be seen as using different strategies to share information. These learning with shared information methods are very effective in solving real-world large-scale problems.}, author = {Darrell, Trevor and Lampert, Christoph and Sebe, Nico and Wu, Ying and Yan, Yan}, journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence}, number = {5}, pages = {1029 -- 1031}, publisher = {IEEE}, title = {{Guest editors' introduction to the special section on learning with Shared information for computer vision and multimedia analysis}}, doi = {10.1109/TPAMI.2018.2804998}, volume = {40}, year = {2018}, } @inproceedings{10882, abstract = {We introduce Intelligent Annotation Dialogs for bounding box annotation. We train an agent to automatically choose a sequence of actions for a human annotator to produce a bounding box in a minimal amount of time. Specifically, we consider two actions: box verification [34], where the annotator verifies a box generated by an object detector, and manual box drawing. We explore two kinds of agents, one based on predicting the probability that a box will be positively verified, and the other based on reinforcement learning. We demonstrate that (1) our agents are able to learn efficient annotation strategies in several scenarios, automatically adapting to the image difficulty, the desired quality of the boxes, and the detector strength; (2) in all scenarios the resulting annotation dialogs speed up annotation compared to manual box drawing alone and box verification alone, while also outperforming any fixed combination of verification and drawing in most scenarios; (3) in a realistic scenario where the detector is iteratively re-trained, our agents evolve a series of strategies that reflect the shifting trade-off between verification and drawing as the detector grows stronger.}, author = {Uijlings, Jasper and Konyushkova, Ksenia and Lampert, Christoph and Ferrari, Vittorio}, booktitle = {2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition}, isbn = {9781538664209}, issn = {2575-7075}, location = {Salt Lake City, UT, United States}, pages = {9175--9184}, publisher = {IEEE}, title = {{Learning intelligent dialogs for bounding box annotation}}, doi = {10.1109/cvpr.2018.00956}, year = {2018}, } @inproceedings{6012, abstract = {We present an approach to identify concise equations from data using a shallow neural network approach. In contrast to ordinary black-box regression, this approach allows understanding functional relations and generalizing them from observed data to unseen parts of the parameter space. We show how to extend the class of learnable equations for a recently proposed equation learning network to include divisions, and we improve the learning and model selection strategy to be useful for challenging real-world data. For systems governed by analytical expressions, our method can in many cases identify the true underlying equation and extrapolate to unseen domains. We demonstrate its effectiveness by experiments on a cart-pendulum system, where only 2 random rollouts are required to learn the forward dynamics and successfully achieve the swing-up task.}, author = {Sahoo, Subham and Lampert, Christoph and Martius, Georg S}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, location = {Stockholm, Sweden}, pages = {4442--4450}, publisher = {ML Research Press}, title = {{Learning equations for extrapolation and control}}, volume = {80}, year = {2018}, } @inproceedings{6011, abstract = {We establish a data-dependent notion of algorithmic stability for Stochastic Gradient Descent (SGD), and employ it to develop novel generalization bounds. This is in contrast to previous distribution-free algorithmic stability results for SGD which depend on the worst-case constants. By virtue of the data-dependent argument, our bounds provide new insights into learning with SGD on convex and non-convex problems. In the convex case, we show that the bound on the generalization error depends on the risk at the initialization point. In the non-convex case, we prove that the expected curvature of the objective function around the initialization point has crucial influence on the generalization error. In both cases, our results suggest a simple data-driven strategy to stabilize SGD by pre-screening its initialization. As a corollary, our results allow us to show optimistic generalization bounds that exhibit fast convergence rates for SGD subject to a vanishing empirical risk and low noise of stochastic gradient. }, author = {Kuzborskij, Ilja and Lampert, Christoph}, booktitle = {Proceedings of the 35 th International Conference on Machine Learning}, location = {Stockholm, Sweden}, pages = {2815--2824}, publisher = {ML Research Press}, title = {{Data-dependent stability of stochastic gradient descent}}, volume = {80}, year = {2018}, } @inproceedings{6589, abstract = {Distributed training of massive machine learning models, in particular deep neural networks, via Stochastic Gradient Descent (SGD) is becoming commonplace. Several families of communication-reduction methods, such as quantization, large-batch methods, and gradient sparsification, have been proposed. To date, gradient sparsification methods--where each node sorts gradients by magnitude, and only communicates a subset of the components, accumulating the rest locally--are known to yield some of the largest practical gains. Such methods can reduce the amount of communication per step by up to \emph{three orders of magnitude}, while preserving model accuracy. Yet, this family of methods currently has no theoretical justification. This is the question we address in this paper. We prove that, under analytic assumptions, sparsifying gradients by magnitude with local error correction provides convergence guarantees, for both convex and non-convex smooth objectives, for data-parallel SGD. The main insight is that sparsification methods implicitly maintain bounds on the maximum impact of stale updates, thanks to selection by magnitude. Our analysis and empirical validation also reveal that these methods do require analytical conditions to converge well, justifying existing heuristics.}, author = {Alistarh, Dan-Adrian and Hoefler, Torsten and Johansson, Mikael and Konstantinov, Nikola H and Khirirat, Sarit and Renggli, Cedric}, booktitle = {Advances in Neural Information Processing Systems 31}, location = {Montreal, Canada}, pages = {5973--5983}, publisher = {Neural Information Processing Systems Foundation}, title = {{The convergence of sparsified gradient methods}}, volume = {Volume 2018}, year = {2018}, } @misc{5584, abstract = {This package contains data for the publication "Nonlinear decoding of a complex movie from the mammalian retina" by Deny S. et al, PLOS Comput Biol (2018). The data consists of (i) 91 spike sorted, isolated rat retinal ganglion cells that pass stability and quality criteria, recorded on the multi-electrode array, in response to the presentation of the complex movie with many randomly moving dark discs. The responses are represented as 648000 x 91 binary matrix, where the first index indicates the timebin of duration 12.5 ms, and the second index the neural identity. The matrix entry is 0/1 if the neuron didn't/did spike in the particular time bin. (ii) README file and a graphical illustration of the structure of the experiment, specifying how the 648000 timebins are split into epochs where 1, 2, 4, or 10 discs were displayed, and which stimulus segments are exact repeats or unique ball trajectories. (iii) a 648000 x 400 matrix of luminance traces for each of the 20 x 20 positions ("sites") in the movie frame, with time that is locked to the recorded raster. The luminance traces are produced as described in the manuscript by filtering the raw disc movie with a small gaussian spatial kernel. }, author = {Deny, Stephane and Marre, Olivier and Botella-Soler, Vicente and Martius, Georg S and Tkacik, Gasper}, keywords = {retina, decoding, regression, neural networks, complex stimulus}, publisher = {Institute of Science and Technology Austria}, title = {{Nonlinear decoding of a complex movie from the mammalian retina}}, doi = {10.15479/AT:ISTA:98}, year = {2018}, } @inproceedings{652, abstract = {We present an approach that enables robots to self-organize their sensorimotor behavior from scratch without providing specific information about neither the robot nor its environment. This is achieved by a simple neural control law that increases the consistency between external sensor dynamics and internal neural dynamics of the utterly simple controller. In this way, the embodiment and the agent-environment coupling are the only source of individual development. We show how an anthropomorphic tendon driven arm-shoulder system develops different behaviors depending on that coupling. For instance: Given a bottle half-filled with water, the arm starts to shake it, driven by the physical response of the water. When attaching a brush, the arm can be manipulated into wiping a table, and when connected to a revolvable wheel it finds out how to rotate it. Thus, the robot may be said to discover the affordances of the world. When allowing two (simulated) humanoid robots to interact physically, they engage into a joint behavior development leading to, for instance, spontaneous cooperation. More social effects are observed if the robots can visually perceive each other. Although, as an observer, it is tempting to attribute an apparent intentionality, there is nothing of the kind put in. As a conclusion, we argue that emergent behavior may be much less rooted in explicit intentions, internal motivations, or specific reward systems than is commonly believed.}, author = {Der, Ralf and Martius, Georg S}, isbn = {978-150905069-7}, location = {Cergy-Pontoise, France}, publisher = {IEEE}, title = {{Dynamical self consistency leads to behavioral development and emergent social interactions in robots}}, doi = {10.1109/DEVLRN.2016.7846789}, year = {2017}, } @article{658, abstract = {With the accelerated development of robot technologies, control becomes one of the central themes of research. In traditional approaches, the controller, by its internal functionality, finds appropriate actions on the basis of specific objectives for the task at hand. While very successful in many applications, self-organized control schemes seem to be favored in large complex systems with unknown dynamics or which are difficult to model. Reasons are the expected scalability, robustness, and resilience of self-organizing systems. The paper presents a self-learning neurocontroller based on extrinsic differential plasticity introduced recently, applying it to an anthropomorphic musculoskeletal robot arm with attached objects of unknown physical dynamics. The central finding of the paper is the following effect: by the mere feedback through the internal dynamics of the object, the robot is learning to relate each of the objects with a very specific sensorimotor pattern. Specifically, an attached pendulum pilots the arm into a circular motion, a half-filled bottle produces axis oriented shaking behavior, a wheel is getting rotated, and wiping patterns emerge automatically in a table-plus-brush setting. By these object-specific dynamical patterns, the robot may be said to recognize the object's identity, or in other words, it discovers dynamical affordances of objects. Furthermore, when including hand coordinates obtained from a camera, a dedicated hand-eye coordination self-organizes spontaneously. These phenomena are discussed from a specific dynamical system perspective. Central is the dedicated working regime at the border to instability with its potentially infinite reservoir of (limit cycle) attractors "waiting" to be excited. Besides converging toward one of these attractors, variate behavior is also arising from a self-induced attractor morphing driven by the learning rule. We claim that experimental investigations with this anthropomorphic, self-learning robot not only generate interesting and potentially useful behaviors, but may also help to better understand what subjective human muscle feelings are, how they can be rooted in sensorimotor patterns, and how these concepts may feed back on robotics.}, author = {Der, Ralf and Martius, Georg S}, issn = {16625218}, journal = {Frontiers in Neurorobotics}, number = {MAR}, publisher = {Frontiers Research Foundation}, title = {{Self organized behavior generation for musculoskeletal robots}}, doi = {10.3389/fnbot.2017.00008}, volume = {11}, year = {2017}, } @inproceedings{6841, abstract = {In classical machine learning, regression is treated as a black box process of identifying a suitable function from a hypothesis set without attempting to gain insight into the mechanism connecting inputs and outputs. In the natural sciences, however, finding an interpretable function for a phenomenon is the prime goal as it allows to understand and generalize results. This paper proposes a novel type of function learning network, called equation learner (EQL), that can learn analytical expressions and is able to extrapolate to unseen domains. It is implemented as an end-to-end differentiable feed-forward network and allows for efficient gradient based training. Due to sparsity regularization concise interpretable expressions can be obtained. Often the true underlying source expression is identified.}, author = {Martius, Georg S and Lampert, Christoph}, booktitle = {5th International Conference on Learning Representations, ICLR 2017 - Workshop Track Proceedings}, location = {Toulon, France}, publisher = {International Conference on Learning Representations}, title = {{Extrapolation and learning equations}}, year = {2017}, } @inproceedings{750, abstract = {Modern communication technologies allow first responders to contact thousands of potential volunteers simultaneously for support during a crisis or disaster event. However, such volunteer efforts must be well coordinated and monitored, in order to offer an effective relief to the professionals. In this paper we extend earlier work on optimally assigning volunteers to selected landmark locations. In particular, we emphasize the aspect that obtaining good assignments requires not only advanced computational tools, but also a realistic measure of distance between volunteers and landmarks. Specifically, we propose the use of the Open Street Map (OSM) driving distance instead of he previously used flight distance. We find the OSM driving distance to be better aligned with the interests of volunteers and first responders. Furthermore, we show that relying on the flying distance leads to a substantial underestimation of the number of required volunteers, causing negative side effects in case of an actual crisis situation.}, author = {Pielorz, Jasmin and Prandtstetter, Matthias and Straub, Markus and Lampert, Christoph}, booktitle = {2017 IEEE International Conference on Big Data}, isbn = {978-153862714-3}, location = {Boston, MA, United States}, pages = {3760 -- 3763}, publisher = {IEEE}, title = {{Optimal geospatial volunteer allocation needs realistic distances}}, doi = {10.1109/BigData.2017.8258375}, year = {2017}, } @inproceedings{1000, abstract = {We study probabilistic models of natural images and extend the autoregressive family of PixelCNN models by incorporating latent variables. Subsequently, we describe two new generative image models that exploit different image transformations as latent variables: a quantized grayscale view of the image or a multi-resolution image pyramid. The proposed models tackle two known shortcomings of existing PixelCNN models: 1) their tendency to focus on low-level image details, while largely ignoring high-level image information, such as object shapes, and 2) their computationally costly procedure for image sampling. We experimentally demonstrate benefits of our LatentPixelCNN models, in particular showing that they produce much more realistically looking image samples than previous state-of-the-art probabilistic models. }, author = {Kolesnikov, Alexander and Lampert, Christoph}, booktitle = {34th International Conference on Machine Learning}, isbn = {978-151085514-4}, location = {Sydney, Australia}, pages = {1905 -- 1914}, publisher = {JMLR}, title = {{PixelCNN models with auxiliary variables for natural image modeling}}, volume = {70}, year = {2017}, } @inproceedings{998, abstract = {A major open problem on the road to artificial intelligence is the development of incrementally learning systems that learn about more and more concepts over time from a stream of data. In this work, we introduce a new training strategy, iCaRL, that allows learning in such a class-incremental way: only the training data for a small number of classes has to be present at the same time and new classes can be added progressively. iCaRL learns strong classifiers and a data representation simultaneously. This distinguishes it from earlier works that were fundamentally limited to fixed data representations and therefore incompatible with deep learning architectures. We show by experiments on CIFAR-100 and ImageNet ILSVRC 2012 data that iCaRL can learn many classes incrementally over a long period of time where other strategies quickly fail. }, author = {Rebuffi, Sylvestre Alvise and Kolesnikov, Alexander and Sperl, Georg and Lampert, Christoph}, isbn = {978-153860457-1}, location = {Honolulu, HA, United States}, pages = {5533 -- 5542}, publisher = {IEEE}, title = {{iCaRL: Incremental classifier and representation learning}}, doi = {10.1109/CVPR.2017.587}, volume = {2017}, year = {2017}, } @inproceedings{911, abstract = {We develop a probabilistic technique for colorizing grayscale natural images. In light of the intrinsic uncertainty of this task, the proposed probabilistic framework has numerous desirable properties. In particular, our model is able to produce multiple plausible and vivid colorizations for a given grayscale image and is one of the first colorization models to provide a proper stochastic sampling scheme. Moreover, our training procedure is supported by a rigorous theoretical framework that does not require any ad hoc heuristics and allows for efficient modeling and learning of the joint pixel color distribution.We demonstrate strong quantitative and qualitative experimental results on the CIFAR-10 dataset and the challenging ILSVRC 2012 dataset.}, author = {Royer, Amélie and Kolesnikov, Alexander and Lampert, Christoph}, location = {London, United Kingdom}, pages = {85.1--85.12}, publisher = {BMVA Press}, title = {{Probabilistic image colorization}}, doi = {10.5244/c.31.85}, year = {2017}, } @inproceedings{1108, abstract = {In this work we study the learnability of stochastic processes with respect to the conditional risk, i.e. the existence of a learning algorithm that improves its next-step performance with the amount of observed data. We introduce a notion of pairwise discrepancy between conditional distributions at different times steps and show how certain properties of these discrepancies can be used to construct a successful learning algorithm. Our main results are two theorems that establish criteria for learnability for many classes of stochastic processes, including all special cases studied previously in the literature.}, author = {Zimin, Alexander and Lampert, Christoph}, location = {Fort Lauderdale, FL, United States}, pages = {213 -- 222}, publisher = {ML Research Press}, title = {{Learning theory for conditional risk minimization}}, volume = {54}, year = {2017}, } @inproceedings{999, abstract = {In multi-task learning, a learner is given a collection of prediction tasks and needs to solve all of them. In contrast to previous work, which required that annotated training data must be available for all tasks, we consider a new setting, in which for some tasks, potentially most of them, only unlabeled training data is provided. Consequently, to solve all tasks, information must be transferred between tasks with labels and tasks without labels. Focusing on an instance-based transfer method we analyze two variants of this setting: when the set of labeled tasks is fixed, and when it can be actively selected by the learner. We state and prove a generalization bound that covers both scenarios and derive from it an algorithm for making the choice of labeled tasks (in the active case) and for transferring information between the tasks in a principled way. We also illustrate the effectiveness of the algorithm on synthetic and real data. }, author = {Pentina, Anastasia and Lampert, Christoph}, isbn = {9781510855144}, location = {Sydney, Australia}, pages = {2807 -- 2816}, publisher = {ML Research Press}, title = {{Multi-task learning with labeled and unlabeled tasks}}, volume = {70}, year = {2017}, } @inproceedings{1098, abstract = {Better understanding of the potential benefits of information transfer and representation learning is an important step towards the goal of building intelligent systems that are able to persist in the world and learn over time. In this work, we consider a setting where the learner encounters a stream of tasks but is able to retain only limited information from each encountered task, such as a learned predictor. In contrast to most previous works analyzing this scenario, we do not make any distributional assumptions on the task generating process. Instead, we formulate a complexity measure that captures the diversity of the observed tasks. We provide a lifelong learning algorithm with error guarantees for every observed task (rather than on average). We show sample complexity reductions in comparison to solving every task in isolation in terms of our task complexity measure. Further, our algorithmic framework can naturally be viewed as learning a representation from encountered tasks with a neural network.}, author = {Pentina, Anastasia and Urner, Ruth}, location = {Barcelona, Spain}, pages = {3619--3627}, publisher = {Neural Information Processing Systems}, title = {{Lifelong learning with weighted majority votes}}, volume = {29}, year = {2016}, } @inproceedings{1102, abstract = {Weakly-supervised object localization methods tend to fail for object classes that consistently co-occur with the same background elements, e.g. trains on tracks. We propose a method to overcome these failures by adding a very small amount of model-specific additional annotation. The main idea is to cluster a deep network\'s mid-level representations and assign object or distractor labels to each cluster. Experiments show substantially improved localization results on the challenging ILSVC2014 dataset for bounding box detection and the PASCAL VOC2012 dataset for semantic segmentation.}, author = {Kolesnikov, Alexander and Lampert, Christoph}, booktitle = {Proceedings of the British Machine Vision Conference 2016}, location = {York, United Kingdom}, pages = {92.1--92.12}, publisher = {BMVA Press}, title = {{Improving weakly-supervised object localization by micro-annotation}}, doi = {10.5244/C.30.92}, volume = {2016-September}, year = {2016}, } @inproceedings{1214, abstract = {With the accelerated development of robot technologies, optimal control becomes one of the central themes of research. In traditional approaches, the controller, by its internal functionality, finds appropriate actions on the basis of the history of sensor values, guided by the goals, intentions, objectives, learning schemes, and so forth. While very successful with classical robots, these methods run into severe difficulties when applied to soft robots, a new field of robotics with large interest for human-robot interaction. We claim that a novel controller paradigm opens new perspective for this field. This paper applies a recently developed neuro controller with differential extrinsic synaptic plasticity to a muscle-tendon driven arm-shoulder system from the Myorobotics toolkit. In the experiments, we observe a vast variety of self-organized behavior patterns: when left alone, the arm realizes pseudo-random sequences of different poses. By applying physical forces, the system can be entrained into definite motion patterns like wiping a table. Most interestingly, after attaching an object, the controller gets in a functional resonance with the object's internal dynamics, starting to shake spontaneously bottles half-filled with water or sensitively driving an attached pendulum into a circular mode. When attached to the crank of a wheel the neural system independently develops to rotate it. In this way, the robot discovers affordances of objects its body is interacting with.}, author = {Martius, Georg S and Hostettler, Raphael and Knoll, Alois and Der, Ralf}, location = {Daejeon, Korea}, publisher = {IEEE}, title = {{Compliant control for soft robots: Emergent behavior of a tendon driven anthropomorphic arm}}, doi = {10.1109/IROS.2016.7759138}, volume = {2016-November}, year = {2016}, } @inproceedings{1369, abstract = {We introduce a new loss function for the weakly-supervised training of semantic image segmentation models based on three guiding principles: to seed with weak localization cues, to expand objects based on the information about which classes can occur in an image, and to constrain the segmentations to coincide with object boundaries. We show experimentally that training a deep convolutional neural network using the proposed loss function leads to substantially better segmentations than previous state-of-the-art methods on the challenging PASCAL VOC 2012 dataset. We furthermore give insight into the working mechanism of our method by a detailed experimental study that illustrates how the segmentation quality is affected by each term of the proposed loss function as well as their combinations.}, author = {Kolesnikov, Alexander and Lampert, Christoph}, location = {Amsterdam, The Netherlands}, pages = {695 -- 711}, publisher = {Springer}, title = {{Seed, expand and constrain: Three principles for weakly-supervised image segmentation}}, doi = {10.1007/978-3-319-46493-0_42}, volume = {9908}, year = {2016}, } @inproceedings{1707, abstract = {Volunteer supporters play an important role in modern crisis and disaster management. In the times of mobile Internet devices, help from thousands of volunteers can be requested within a short time span, thus relieving professional helpers from minor chores or geographically spread-out tasks. However, the simultaneous availability of many volunteers also poses new problems. In particular, the volunteer efforts must be well coordinated, or otherwise situations might emerge in which too many idle volunteers at one location become more of a burden than a relief to the professionals. In this work, we study the task of optimally assigning volunteers to selected locations, e.g. in order to perform regular measurements, to report on damage, or to distribute information or resources to the population in a crisis situation. We formulate the assignment tasks as an optimization problem and propose an effective and efficient solution procedure. Experiments on real data of the Team Österreich, consisting of over 36,000 Austrian volunteers, show the effectiveness and efficiency of our approach.}, author = {Pielorz, Jasmin and Lampert, Christoph}, location = {Rennes, France}, publisher = {IEEE}, title = {{Optimal geospatial allocation of volunteers for crisis management}}, doi = {10.1109/ICT-DM.2015.7402041}, year = {2016}, } @inproceedings{8094, abstract = {With the accelerated development of robot technologies, optimal control becomes one of the central themes of research. In traditional approaches, the controller, by its internal functionality, finds appropriate actions on the basis of the history of sensor values, guided by the goals, intentions, objectives, learning schemes, and so forth. The idea is that the controller controls the world---the body plus its environment---as reliably as possible. This paper focuses on new lines of self-organization for developmental robotics. We apply the recently developed differential extrinsic synaptic plasticity to a muscle-tendon driven arm-shoulder system from the Myorobotics toolkit. In the experiments, we observe a vast variety of self-organized behavior patterns: when left alone, the arm realizes pseudo-random sequences of different poses. By applying physical forces, the system can be entrained into definite motion patterns like wiping a table. Most interestingly, after attaching an object, the controller gets in a functional resonance with the object's internal dynamics, starting to shake spontaneously bottles half-filled with water or sensitively driving an attached pendulum into a circular mode. When attached to the crank of a wheel the neural system independently discovers how to rotate it. In this way, the robot discovers affordances of objects its body is interacting with.}, author = {Martius, Georg S and Hostettler, Rafael and Knoll, Alois and Der, Ralf}, booktitle = {Proceedings of the Artificial Life Conference 2016}, isbn = {9780262339360}, location = {Cancun, Mexico}, pages = {142--143}, publisher = {MIT Press}, title = {{Self-organized control of an tendon driven arm by differential extrinsic plasticity}}, doi = {10.7551/978-0-262-33936-0-ch029}, volume = {28}, year = {2016}, } @phdthesis{1126, abstract = {Traditionally machine learning has been focusing on the problem of solving a single task in isolation. While being quite well understood, this approach disregards an important aspect of human learning: when facing a new problem, humans are able to exploit knowledge acquired from previously learned tasks. Intuitively, access to several problems simultaneously or sequentially could also be advantageous for a machine learning system, especially if these tasks are closely related. Indeed, results of many empirical studies have provided justification for this intuition. However, theoretical justifications of this idea are rather limited. The focus of this thesis is to expand the understanding of potential benefits of information transfer between several related learning problems. We provide theoretical analysis for three scenarios of multi-task learning - multiple kernel learning, sequential learning and active task selection. We also provide a PAC-Bayesian perspective on lifelong learning and investigate how the task generation process influences the generalization guarantees in this scenario. In addition, we show how some of the obtained theoretical results can be used to derive principled multi-task and lifelong learning algorithms and illustrate their performance on various synthetic and real-world datasets.}, author = {Pentina, Anastasia}, issn = {2663-337X}, pages = {127}, publisher = {Institute of Science and Technology Austria}, title = {{Theoretical foundations of multi-task lifelong learning}}, doi = {10.15479/AT:ISTA:TH_776}, year = {2016}, } @inproceedings{1425, abstract = {In this work we aim at extending the theoretical foundations of lifelong learning. Previous work analyzing this scenario is based on the assumption that learning tasks are sampled i.i.d. from a task environment or limited to strongly constrained data distributions. Instead, we study two scenarios when lifelong learning is possible, even though the observed tasks do not form an i.i.d. sample: first, when they are sampled from the same environment, but possibly with dependencies, and second, when the task environment is allowed to change over time in a consistent way. In the first case we prove a PAC-Bayesian theorem that can be seen as a direct generalization of the analogous previous result for the i.i.d. case. For the second scenario we propose to learn an inductive bias in form of a transfer procedure. We present a generalization bound and show on a toy example how it can be used to identify a beneficial transfer algorithm.}, author = {Pentina, Anastasia and Lampert, Christoph}, location = {Montreal, Canada}, pages = {1540 -- 1548}, publisher = {Neural Information Processing Systems}, title = {{Lifelong learning with non-i.i.d. tasks}}, volume = {2015}, year = {2015}, } @article{1533, abstract = {This paper addresses the problem of semantic segmentation, where the possible class labels are from a predefined set. We exploit top-down guidance, i.e., the coarse localization of the objects and their class labels provided by object detectors. For each detected bounding box, figure-ground segmentation is performed and the final result is achieved by merging the figure-ground segmentations. The main idea of the proposed approach, which is presented in our preliminary work, is to reformulate the figure-ground segmentation problem as sparse reconstruction pursuing the object mask in a nonparametric manner. The latent segmentation mask should be coherent subject to sparse error caused by intra-category diversity; thus, the object mask is inferred by making use of sparse representations over the training set. To handle local spatial deformations, local patch-level masks are also considered and inferred by sparse representations over the spatially nearby patches. The sparse reconstruction coefficients and the latent mask are alternately optimized by applying the Lasso algorithm and the accelerated proximal gradient method. The proposed formulation results in a convex optimization problem; thus, the global optimal solution is achieved. In this paper, we provide theoretical analysis of the convergence and optimality. We also give an extended numerical analysis of the proposed algorithm and a comprehensive comparison with the related semantic segmentation methods on the challenging PASCAL visual object class object segmentation datasets and the Weizmann horse dataset. The experimental results demonstrate that the proposed algorithm achieves a competitive performance when compared with the state of the arts.}, author = {Xia, Wei and Domokos, Csaba and Xiong, Junjun and Cheong, Loongfah and Yan, Shuicheng}, journal = {IEEE Transactions on Circuits and Systems for Video Technology}, number = {8}, pages = {1295 -- 1308}, publisher = {IEEE}, title = {{Segmentation over detection via optimal sparse reconstructions}}, doi = {10.1109/TCSVT.2014.2379972}, volume = {25}, year = {2015}, } @article{1570, abstract = {Grounding autonomous behavior in the nervous system is a fundamental challenge for neuroscience. In particular, self-organized behavioral development provides more questions than answers. Are there special functional units for curiosity, motivation, and creativity? This paper argues that these features can be grounded in synaptic plasticity itself, without requiring any higher-level constructs. We propose differential extrinsic plasticity (DEP) as a new synaptic rule for self-learning systems and apply it to a number of complex robotic systems as a test case. Without specifying any purpose or goal, seemingly purposeful and adaptive rhythmic behavior is developed, displaying a certain level of sensorimotor intelligence. These surprising results require no systemspecific modifications of the DEP rule. They rather arise from the underlying mechanism of spontaneous symmetry breaking,which is due to the tight brain body environment coupling. The new synaptic rule is biologically plausible and would be an interesting target for neurobiological investigation. We also argue that this neuronal mechanism may have been a catalyst in natural evolution.}, author = {Der, Ralf and Martius, Georg S}, journal = {PNAS}, number = {45}, pages = {E6224 -- E6232}, publisher = {National Academy of Sciences}, title = {{Novel plasticity rule can explain the development of sensorimotor intelligence}}, doi = {10.1073/pnas.1508400112}, volume = {112}, year = {2015}, } @inproceedings{1706, abstract = {We consider a problem of learning kernels for use in SVM classification in the multi-task and lifelong scenarios and provide generalization bounds on the error of a large margin classifier. Our results show that, under mild conditions on the family of kernels used for learning, solving several related tasks simultaneously is beneficial over single task learning. In particular, as the number of observed tasks grows, assuming that in the considered family of kernels there exists one that yields low approximation error on all tasks, the overhead associated with learning such a kernel vanishes and the complexity converges to that of learning when this good kernel is given to the learner.}, author = {Pentina, Anastasia and Ben David, Shai}, location = {Banff, AB, Canada}, pages = {194 -- 208}, publisher = {Springer}, title = {{Multi-task and lifelong learning of kernels}}, doi = {10.1007/978-3-319-24486-0_13}, volume = {9355}, year = {2015}, } @inproceedings{1859, abstract = {Structural support vector machines (SSVMs) are amongst the best performing models for structured computer vision tasks, such as semantic image segmentation or human pose estimation. Training SSVMs, however, is computationally costly, because it requires repeated calls to a structured prediction subroutine (called \emph{max-oracle}), which has to solve an optimization problem itself, e.g. a graph cut. In this work, we introduce a new algorithm for SSVM training that is more efficient than earlier techniques when the max-oracle is computationally expensive, as it is frequently the case in computer vision tasks. The main idea is to (i) combine the recent stochastic Block-Coordinate Frank-Wolfe algorithm with efficient hyperplane caching, and (ii) use an automatic selection rule for deciding whether to call the exact max-oracle or to rely on an approximate one based on the cached hyperplanes. We show experimentally that this strategy leads to faster convergence to the optimum with respect to the number of requires oracle calls, and that this translates into faster convergence with respect to the total runtime when the max-oracle is slow compared to the other steps of the algorithm. }, author = {Shah, Neel and Kolmogorov, Vladimir and Lampert, Christoph}, location = {Boston, MA, USA}, pages = {2737 -- 2745}, publisher = {IEEE}, title = {{A multi-plane block-coordinate Frank-Wolfe algorithm for training structural SVMs with a costly max-oracle}}, doi = {10.1109/CVPR.2015.7298890}, year = {2015}, } @inproceedings{1860, abstract = {Classifiers for object categorization are usually evaluated by their accuracy on a set of i.i.d. test examples. This provides us with an estimate of the expected error when applying the classifiers to a single new image. In real application, however, classifiers are rarely only used for a single image and then discarded. Instead, they are applied sequentially to many images, and these are typically not i.i.d. samples from a fixed data distribution, but they carry dependencies and their class distribution varies over time. In this work, we argue that the phenomenon of correlated data at prediction time is not a nuisance, but a blessing in disguise. We describe a probabilistic method for adapting classifiers at prediction time without having to retrain them. We also introduce a framework for creating realistically distributed image sequences, which offers a way to benchmark classifier adaptation methods, such as the one we propose. Experiments on the ILSVRC2010 and ILSVRC2012 datasets show that adapting object classification systems at prediction time can significantly reduce their error rate, even with no additional human feedback.}, author = {Royer, Amélie and Lampert, Christoph}, location = {Boston, MA, United States}, pages = {1401 -- 1409}, publisher = {IEEE}, title = {{Classifier adaptation at prediction time}}, doi = {10.1109/CVPR.2015.7298746}, year = {2015}, } @inproceedings{1858, abstract = {We study the problem of predicting the future, though only in the probabilistic sense of estimating a future state of a time-varying probability distribution. This is not only an interesting academic problem, but solving this extrapolation problem also has many practical application, e.g. for training classifiers that have to operate under time-varying conditions. Our main contribution is a method for predicting the next step of the time-varying distribution from a given sequence of sample sets from earlier time steps. For this we rely on two recent machine learning techniques: embedding probability distributions into a reproducing kernel Hilbert space, and learning operators by vector-valued regression. We illustrate the working principles and the practical usefulness of our method by experiments on synthetic and real data. We also highlight an exemplary application: training a classifier in a domain adaptation setting without having access to examples from the test time distribution at training time.}, author = {Lampert, Christoph}, location = {Boston, MA, United States}, pages = {942 -- 950}, publisher = {IEEE}, title = {{Predicting the future behavior of a time-varying probability distribution}}, doi = {10.1109/CVPR.2015.7298696}, year = {2015}, } @inproceedings{1857, abstract = {Sharing information between multiple tasks enables algorithms to achieve good generalization performance even from small amounts of training data. However, in a realistic scenario of multi-task learning not all tasks are equally related to each other, hence it could be advantageous to transfer information only between the most related tasks. In this work we propose an approach that processes multiple tasks in a sequence with sharing between subsequent tasks instead of solving all tasks jointly. Subsequently, we address the question of curriculum learning of tasks, i.e. finding the best order of tasks to be learned. Our approach is based on a generalization bound criterion for choosing the task order that optimizes the average expected classification performance over all tasks. Our experimental results show that learning multiple related tasks sequentially can be more effective than learning them jointly, the order in which tasks are being solved affects the overall performance, and that our model is able to automatically discover the favourable order of tasks. }, author = {Pentina, Anastasia and Sharmanska, Viktoriia and Lampert, Christoph}, location = {Boston, MA, United States}, pages = {5492 -- 5500}, publisher = {IEEE}, title = {{Curriculum learning of multiple tasks}}, doi = {10.1109/CVPR.2015.7299188}, year = {2015}, } @inproceedings{12881, author = {Martius, Georg S and Olbrich, Eckehard}, booktitle = {Proceedings of the 13th European Conference on Artificial Life}, isbn = {9780262330275}, location = {York, United Kingdom}, pages = {78}, publisher = {MIT Press}, title = {{Quantifying self-organizing behavior of autonomous robots}}, doi = {10.7551/978-0-262-33027-5-ch018}, year = {2015}, } @phdthesis{1401, abstract = {The human ability to recognize objects in complex scenes has driven research in the computer vision field over couple of decades. This thesis focuses on the object recognition task in images. That is, given the image, we want the computer system to be able to predict the class of the object that appears in the image. A recent successful attempt to bridge semantic understanding of the image perceived by humans and by computers uses attribute-based models. Attributes are semantic properties of the objects shared across different categories, which humans and computers can decide on. To explore the attribute-based models we take a statistical machine learning approach, and address two key learning challenges in view of object recognition task: learning augmented attributes as mid-level discriminative feature representation, and learning with attributes as privileged information. Our main contributions are parametric and non-parametric models and algorithms to solve these frameworks. In the parametric approach, we explore an autoencoder model combined with the large margin nearest neighbor principle for mid-level feature learning, and linear support vector machines for learning with privileged information. In the non-parametric approach, we propose a supervised Indian Buffet Process for automatic augmentation of semantic attributes, and explore the Gaussian Processes classification framework for learning with privileged information. A thorough experimental analysis shows the effectiveness of the proposed models in both parametric and non-parametric views.}, author = {Sharmanska, Viktoriia}, issn = {2663-337X}, pages = {144}, publisher = {Institute of Science and Technology Austria}, title = {{Learning with attributes for object recognition: Parametric and non-parametrics views}}, doi = {10.15479/at:ista:1401}, year = {2015}, } @article{1655, abstract = {Quantifying behaviors of robots which were generated autonomously from task-independent objective functions is an important prerequisite for objective comparisons of algorithms and movements of animals. The temporal sequence of such a behavior can be considered as a time series and hence complexity measures developed for time series are natural candidates for its quantification. The predictive information and the excess entropy are such complexity measures. They measure the amount of information the past contains about the future and thus quantify the nonrandom structure in the temporal sequence. However, when using these measures for systems with continuous states one has to deal with the fact that their values will depend on the resolution with which the systems states are observed. For deterministic systems both measures will diverge with increasing resolution. We therefore propose a new decomposition of the excess entropy in resolution dependent and resolution independent parts and discuss how they depend on the dimensionality of the dynamics, correlations and the noise level. For the practical estimation we propose to use estimates based on the correlation integral instead of the direct estimation of the mutual information based on next neighbor statistics because the latter allows less control of the scale dependencies. Using our algorithm we are able to show how autonomous learning generates behavior of increasing complexity with increasing learning duration.}, author = {Martius, Georg S and Olbrich, Eckehard}, journal = {Entropy}, number = {10}, pages = {7266 -- 7297}, publisher = {MDPI}, title = {{Quantifying emergent behavior of autonomous robots}}, doi = {10.3390/e17107266}, volume = {17}, year = {2015}, } @inbook{1829, abstract = {Hitting and batting tasks, such as tennis forehands, ping-pong strokes, or baseball batting, depend on predictions where the ball can be intercepted and how it can properly be returned to the opponent. These predictions get more accurate over time, hence the behaviors need to be continuously modified. As a result, movement templates with a learned global shape need to be adapted during the execution so that the racket reaches a target position and velocity that will return the ball over to the other side of the net or court. It requires altering learned movements to hit a varying target with the necessary velocity at a specific instant in time. Such a task cannot be incorporated straightforwardly in most movement representations suitable for learning. For example, the standard formulation of the dynamical system based motor primitives (introduced by Ijspeert et al (2002b)) does not satisfy this property despite their flexibility which has allowed learning tasks ranging from locomotion to kendama. In order to fulfill this requirement, we reformulate the Ijspeert framework to incorporate the possibility of specifying a desired hitting point and a desired hitting velocity while maintaining all advantages of the original formulation.We show that the proposed movement template formulation works well in two scenarios, i.e., for hitting a ball on a string with a table tennis racket at a specified velocity and for returning balls launched by a ball gun successfully over the net using forehand movements.}, author = {Muelling, Katharina and Kroemer, Oliver and Lampert, Christoph and Schölkopf, Bernhard}, booktitle = {Learning Motor Skills}, editor = {Kober, Jens and Peters, Jan}, pages = {69 -- 82}, publisher = {Springer}, title = {{Movement templates for learning of hitting and batting}}, doi = {10.1007/978-3-319-03194-1_3}, volume = {97}, year = {2014}, } @inproceedings{2033, abstract = {The learning with privileged information setting has recently attracted a lot of attention within the machine learning community, as it allows the integration of additional knowledge into the training process of a classifier, even when this comes in the form of a data modality that is not available at test time. Here, we show that privileged information can naturally be treated as noise in the latent function of a Gaussian process classifier (GPC). That is, in contrast to the standard GPC setting, the latent function is not just a nuisance but a feature: it becomes a natural measure of confidence about the training data by modulating the slope of the GPC probit likelihood function. Extensive experiments on public datasets show that the proposed GPC method using privileged noise, called GPC+, improves over a standard GPC without privileged knowledge, and also over the current state-of-the-art SVM-based method, SVM+. Moreover, we show that advanced neural networks and deep learning methods can be compressed as privileged information.}, author = {Hernandez Lobato, Daniel and Sharmanska, Viktoriia and Kersting, Kristian and Lampert, Christoph and Quadrianto, Novi}, booktitle = {Advances in Neural Information Processing Systems}, location = {Montreal, Canada}, number = {January}, pages = {837--845}, publisher = {Neural Information Processing Systems}, title = {{Mind the nuisance: Gaussian process classification using privileged noise}}, volume = {1}, year = {2014}, } @inproceedings{2057, abstract = {In the past few years, a lot of attention has been devoted to multimedia indexing by fusing multimodal informations. Two kinds of fusion schemes are generally considered: The early fusion and the late fusion. We focus on late classifier fusion, where one combines the scores of each modality at the decision level. To tackle this problem, we investigate a recent and elegant well-founded quadratic program named MinCq coming from the machine learning PAC-Bayesian theory. MinCq looks for the weighted combination, over a set of real-valued functions seen as voters, leading to the lowest misclassification rate, while maximizing the voters’ diversity. We propose an extension of MinCq tailored to multimedia indexing. Our method is based on an order-preserving pairwise loss adapted to ranking that allows us to improve Mean Averaged Precision measure while taking into account the diversity of the voters that we want to fuse. We provide evidence that this method is naturally adapted to late fusion procedures and confirm the good behavior of our approach on the challenging PASCAL VOC’07 benchmark.}, author = {Morvant, Emilie and Habrard, Amaury and Ayache, Stéphane}, booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}, location = {Joensuu, Finland}, pages = {153 -- 162}, publisher = {Springer}, title = {{Majority vote of diverse classifiers for late fusion}}, doi = {10.1007/978-3-662-44415-3_16}, volume = {8621}, year = {2014}, } @inproceedings{2171, abstract = {We present LS-CRF, a new method for training cyclic Conditional Random Fields (CRFs) from large datasets that is inspired by classical closed-form expressions for the maximum likelihood parameters of a generative graphical model with tree topology. Training a CRF with LS-CRF requires only solving a set of independent regression problems, each of which can be solved efficiently in closed form or by an iterative solver. This makes LS-CRF orders of magnitude faster than classical CRF training based on probabilistic inference, and at the same time more flexible and easier to implement than other approximate techniques, such as pseudolikelihood or piecewise training. We apply LS-CRF to the task of semantic image segmentation, showing that it achieves on par accuracy to other training techniques at higher speed, thereby allowing efficient CRF training from very large training sets. For example, training a linearly parameterized pairwise CRF on 150,000 images requires less than one hour on a modern workstation.}, author = {Kolesnikov, Alexander and Guillaumin, Matthieu and Ferrari, Vittorio and Lampert, Christoph}, booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}, editor = {Fleet, David and Pajdla, Tomas and Schiele, Bernt and Tuytelaars, Tinne}, location = {Zurich, Switzerland}, number = {PART 3}, pages = {550 -- 565}, publisher = {Springer}, title = {{Closed-form approximate CRF training for scalable image segmentation}}, doi = {10.1007/978-3-319-10578-9_36}, volume = {8691}, year = {2014}, } @inproceedings{2173, abstract = {In this work we introduce a new approach to co-classification, i.e. the task of jointly classifying multiple, otherwise independent, data samples. The method we present, named CoConut, is based on the idea of adding a regularizer in the label space to encode certain priors on the resulting labelings. A regularizer that encourages labelings that are smooth across the test set, for instance, can be seen as a test-time variant of the cluster assumption, which has been proven useful at training time in semi-supervised learning. A regularizer that introduces a preference for certain class proportions can be regarded as a prior distribution on the class labels. CoConut can build on existing classifiers without making any assumptions on how they were obtained and without the need to re-train them. The use of a regularizer adds a new level of flexibility. It allows the integration of potentially new information at test time, even in other modalities than what the classifiers were trained on. We evaluate our framework on six datasets, reporting a clear performance gain in classification accuracy compared to the standard classification setup that predicts labels for each test sample separately. }, author = {Khamis, Sameh and Lampert, Christoph}, booktitle = {Proceedings of the British Machine Vision Conference 2014}, location = {Nottingham, UK}, publisher = {BMVA Press}, title = {{CoConut: Co-classification with output space regularization}}, year = {2014}, } @inproceedings{2172, abstract = {Fisher Kernels and Deep Learning were two developments with significant impact on large-scale object categorization in the last years. Both approaches were shown to achieve state-of-the-art results on large-scale object categorization datasets, such as ImageNet. Conceptually, however, they are perceived as very different and it is not uncommon for heated debates to spring up when advocates of both paradigms meet at conferences or workshops. In this work, we emphasize the similarities between both architectures rather than their differences and we argue that such a unified view allows us to transfer ideas from one domain to the other. As a concrete example we introduce a method for learning a support vector machine classifier with Fisher kernel at the same time as a task-specific data representation. We reinterpret the setting as a multi-layer feed forward network. Its final layer is the classifier, parameterized by a weight vector, and the two previous layers compute Fisher vectors, parameterized by the coefficients of a Gaussian mixture model. We introduce a gradient descent based learning algorithm that, in contrast to other feature learning techniques, is not just derived from intuition or biological analogy, but has a theoretical justification in the framework of statistical learning theory. Our experiments show that the new training procedure leads to significant improvements in classification accuracy while preserving the modularity and geometric interpretability of a support vector machine setup.}, author = {Sydorov, Vladyslav and Sakurada, Mayu and Lampert, Christoph}, booktitle = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}, location = {Columbus, USA}, pages = {1402 -- 1409}, publisher = {IEEE}, title = {{Deep Fisher Kernels – End to end learning of the Fisher Kernel GMM parameters}}, doi = {10.1109/CVPR.2014.182}, year = {2014}, } @article{2180, abstract = {Weighted majority votes allow one to combine the output of several classifiers or voters. MinCq is a recent algorithm for optimizing the weight of each voter based on the minimization of a theoretical bound over the risk of the vote with elegant PAC-Bayesian generalization guarantees. However, while it has demonstrated good performance when combining weak classifiers, MinCq cannot make use of the useful a priori knowledge that one may have when using a mixture of weak and strong voters. In this paper, we propose P-MinCq, an extension of MinCq that can incorporate such knowledge in the form of a constraint over the distribution of the weights, along with general proofs of convergence that stand in the sample compression setting for data-dependent voters. The approach is applied to a vote of k-NN classifiers with a specific modeling of the voters' performance. P-MinCq significantly outperforms the classic k-NN classifier, a symmetric NN and MinCq using the same voters. We show that it is also competitive with LMNN, a popular metric learning algorithm, and that combining both approaches further reduces the error.}, author = {Bellet, Aurélien and Habrard, Amaury and Morvant, Emilie and Sebban, Marc}, journal = {Machine Learning}, number = {1-2}, pages = {129 -- 154}, publisher = {Springer}, title = {{Learning a priori constrained weighted majority votes}}, doi = {10.1007/s10994-014-5462-z}, volume = {97}, year = {2014}, } @inproceedings{2189, abstract = {En apprentissage automatique, nous parlons d'adaptation de domaine lorsque les données de test (cibles) et d'apprentissage (sources) sont générées selon différentes distributions. Nous devons donc développer des algorithmes de classification capables de s'adapter à une nouvelle distribution, pour laquelle aucune information sur les étiquettes n'est disponible. Nous attaquons cette problématique sous l'angle de l'approche PAC-Bayésienne qui se focalise sur l'apprentissage de modèles définis comme des votes de majorité sur un ensemble de fonctions. Dans ce contexte, nous introduisons PV-MinCq une version adaptative de l'algorithme (non adaptatif) MinCq. PV-MinCq suit le principe suivant. Nous transférons les étiquettes sources aux points cibles proches pour ensuite appliquer MinCq sur l'échantillon cible ``auto-étiqueté'' (justifié par une borne théorique). Plus précisément, nous définissons un auto-étiquetage non itératif qui se focalise dans les régions où les distributions marginales source et cible sont les plus similaires. Dans un second temps, nous étudions l'influence de notre auto-étiquetage pour en déduire une procédure de validation des hyperparamètres. Finalement, notre approche montre des résultats empiriques prometteurs.}, author = {Morvant, Emilie}, location = {Saint-Etienne, France}, pages = {49--58}, publisher = {Elsevier}, title = {{Adaptation de domaine de vote de majorité par auto-étiquetage non itératif}}, volume = {1}, year = {2014}, } @inproceedings{2160, abstract = {Transfer learning has received a lot of attention in the machine learning community over the last years, and several effective algorithms have been developed. However, relatively little is known about their theoretical properties, especially in the setting of lifelong learning, where the goal is to transfer information to tasks for which no data have been observed so far. In this work we study lifelong learning from a theoretical perspective. Our main result is a PAC-Bayesian generalization bound that offers a unified view on existing paradigms for transfer learning, such as the transfer of parameters or the transfer of low-dimensional representations. We also use the bound to derive two principled lifelong learning algorithms, and we show that these yield results comparable with existing methods.}, author = {Pentina, Anastasia and Lampert, Christoph}, location = {Beijing, China}, pages = {991 -- 999}, publisher = {ML Research Press}, title = {{A PAC-Bayesian bound for Lifelong Learning}}, volume = {32}, year = {2014}, } @inproceedings{2294, abstract = {In this work we propose a system for automatic classification of Drosophila embryos into developmental stages. While the system is designed to solve an actual problem in biological research, we believe that the principle underly- ing it is interesting not only for biologists, but also for researchers in computer vision. The main idea is to combine two orthogonal sources of information: one is a classifier trained on strongly invariant features, which makes it applicable to images of very different conditions, but also leads to rather noisy predictions. The other is a label propagation step based on a more powerful similarity measure that however is only consistent within specific subsets of the data at a time. In our biological setup, the information sources are the shape and the staining patterns of embryo images. We show experimentally that while neither of the methods can be used by itself to achieve satisfactory results, their combina- tion achieves prediction quality comparable to human performance.}, author = {Kazmar, Tomas and Kvon, Evgeny and Stark, Alexander and Lampert, Christoph}, location = {Sydney, Australia}, publisher = {IEEE}, title = {{Drosophila Embryo Stage Annotation using Label Propagation}}, doi = {10.1109/ICCV.2013.139}, year = {2013}, } @inproceedings{2293, abstract = {Many computer vision problems have an asymmetric distribution of information between training and test time. In this work, we study the case where we are given additional information about the training data, which however will not be available at test time. This situation is called learning using privileged information (LUPI). We introduce two maximum-margin techniques that are able to make use of this additional source of information, and we show that the framework is applicable to several scenarios that have been studied in computer vision before. Experiments with attributes, bounding boxes, image tags and rationales as additional information in object classification show promising results.}, author = {Sharmanska, Viktoriia and Quadrianto, Novi and Lampert, Christoph}, location = {Sydney, Australia}, pages = {825 -- 832}, publisher = {IEEE}, title = {{Learning to rank using privileged information}}, doi = {10.1109/ICCV.2013.107}, year = {2013}, } @article{2516, abstract = {We study the problem of object recognition for categories for which we have no training examples, a task also called zero-data or zero-shot learning. This situation has hardly been studied in computer vision research, even though it occurs frequently: the world contains tens of thousands of different object classes and for only few of them image collections have been formed and suitably annotated. To tackle the problem we introduce attribute-based classification: objects are identified based on a high-level description that is phrased in terms of semantic attributes, such as the object's color or shape. Because the identification of each such property transcends the specific learning task at hand, the attribute classifiers can be pre-learned independently, e.g. from existing image datasets unrelated to the current task. Afterwards, new classes can be detected based on their attribute representation, without the need for a new training phase. In this paper we also introduce a new dataset, Animals with Attributes, of over 30,000 images of 50 animal classes, annotated with 85 semantic attributes. Extensive experiments on this and two more datasets show that attribute-based classification indeed is able to categorize images without access to any training images of the target classes.}, author = {Lampert, Christoph and Nickisch, Hannes and Harmeling, Stefan}, journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence}, number = {3}, pages = {453 -- 465}, publisher = {IEEE}, title = {{Attribute-based classification for zero-shot learning of object categories}}, doi = {10.1109/TPAMI.2013.140}, volume = {36}, year = {2013}, } @inproceedings{2520, abstract = {We propose a probabilistic model to infer supervised latent variables in the Hamming space from observed data. Our model allows simultaneous inference of the number of binary latent variables, and their values. The latent variables preserve neighbourhood structure of the data in a sense that objects in the same semantic concept have similar latent values, and objects in different concepts have dissimilar latent values. We formulate the supervised infinite latent variable problem based on an intuitive principle of pulling objects together if they are of the same type, and pushing them apart if they are not. We then combine this principle with a flexible Indian Buffet Process prior on the latent variables. We show that the inferred supervised latent variables can be directly used to perform a nearest neighbour search for the purpose of retrieval. We introduce a new application of dynamically extending hash codes, and show how to effectively couple the structure of the hash codes with continuously growing structure of the neighbourhood preserving infinite latent feature space.}, author = {Quadrianto, Novi and Sharmanska, Viktoriia and Knowles, David and Ghahramani, Zoubin}, booktitle = {Proceedings of the 29th conference uncertainty in Artificial Intelligence}, isbn = {9780974903996}, location = {Bellevue, WA, United States}, pages = {527 -- 536}, publisher = {AUAI Press}, title = {{The supervised IBP: Neighbourhood preserving infinite latent feature models}}, year = {2013}, } @inproceedings{2901, abstract = { We introduce the M-modes problem for graphical models: predicting the M label configurations of highest probability that are at the same time local maxima of the probability landscape. M-modes have multiple possible applications: because they are intrinsically diverse, they provide a principled alternative to non-maximum suppression techniques for structured prediction, they can act as codebook vectors for quantizing the configuration space, or they can form component centers for mixture model approximation. We present two algorithms for solving the M-modes problem. The first algorithm solves the problem in polynomial time when the underlying graphical model is a simple chain. The second algorithm solves the problem for junction chains. In synthetic and real dataset, we demonstrate how M-modes can improve the performance of prediction. We also use the generated modes as a tool to understand the topography of the probability distribution of configurations, for example with relation to the training set size and amount of noise in the data. }, author = {Chen, Chao and Kolmogorov, Vladimir and Yan, Zhu and Metaxas, Dimitris and Lampert, Christoph}, location = {Scottsdale, AZ, United States}, pages = {161 -- 169}, publisher = {JMLR}, title = {{Computing the M most probable modes of a graphical model}}, volume = {31}, year = {2013}, } @inproceedings{2948, abstract = {Many visual datasets are traditionally used to analyze the performance of different learning techniques. The evaluation is usually done within each dataset, therefore it is questionable if such results are a reliable indicator of true generalization ability. We propose here an algorithm to exploit the existing data resources when learning on a new multiclass problem. Our main idea is to identify an image representation that decomposes orthogonally into two subspaces: a part specific to each dataset, and a part generic to, and therefore shared between, all the considered source sets. This allows us to use the generic representation as un-biased reference knowledge for a novel classification task. By casting the method in the multi-view setting, we also make it possible to use different features for different databases. We call the algorithm MUST, Multitask Unaligned Shared knowledge Transfer. Through extensive experiments on five public datasets, we show that MUST consistently improves the cross-datasets generalization performance.}, author = {Tommasi, Tatiana and Quadrianto, Novi and Caputo, Barbara and Lampert, Christoph}, location = {Daejeon, Korea}, pages = {1 -- 15}, publisher = {Springer}, title = {{Beyond dataset bias: Multi-task unaligned shared knowledge transfer}}, doi = {10.1007/978-3-642-37331-2_1}, volume = {7724}, year = {2013}, } @misc{3321, author = {Quadrianto, Novi and Lampert, Christoph}, booktitle = {Encyclopedia of Systems Biology}, editor = {Dubitzky, Werner and Wolkenhauer, Olaf and Cho, Kwang and Yokota, Hiroki}, pages = {1069 -- 1069}, publisher = {Springer}, title = {{Kernel based learning}}, doi = {10.1007/978-1-4419-9863-7_604}, volume = {3}, year = {2013}, } @inproceedings{2825, abstract = {We study the problem of maximum marginal prediction (MMP) in probabilistic graphical models, a task that occurs, for example, as the Bayes optimal decision rule under a Hamming loss. MMP is typically performed as a two-stage procedure: one estimates each variable's marginal probability and then forms a prediction from the states of maximal probability. In this work we propose a simple yet effective technique for accelerating MMP when inference is sampling-based: instead of the above two-stage procedure we directly estimate the posterior probability of each decision variable. This allows us to identify the point of time when we are sufficiently certain about any individual decision. Whenever this is the case, we dynamically prune the variables we are confident about from the underlying factor graph. Consequently, at any time only samples of variables whose decision is still uncertain need to be created. Experiments in two prototypical scenarios, multi-label classification and image inpainting, show that adaptive sampling can drastically accelerate MMP without sacrificing prediction accuracy.}, author = {Lampert, Christoph}, location = {Lake Tahoe, NV, United States}, pages = {82 -- 90}, publisher = {Neural Information Processing Systems}, title = {{Dynamic pruning of factor graphs for maximum marginal prediction}}, volume = {1}, year = {2012}, } @article{3164, abstract = {Overview of the Special Issue on structured prediction and inference.}, author = {Blaschko, Matthew and Lampert, Christoph}, journal = {International Journal of Computer Vision}, number = {3}, pages = {257 -- 258}, publisher = {Springer}, title = {{Guest editorial: Special issue on structured prediction and inference}}, doi = {10.1007/s11263-012-0530-y}, volume = {99}, year = {2012}, } @inproceedings{3125, abstract = {We propose a new learning method to infer a mid-level feature representation that combines the advantage of semantic attribute representations with the higher expressive power of non-semantic features. The idea lies in augmenting an existing attribute-based representation with additional dimensions for which an autoencoder model is coupled with a large-margin principle. This construction allows a smooth transition between the zero-shot regime with no training example, the unsupervised regime with training examples but without class labels, and the supervised regime with training examples and with class labels. The resulting optimization problem can be solved efficiently, because several of the necessity steps have closed-form solutions. Through extensive experiments we show that the augmented representation achieves better results in terms of object categorization accuracy than the semantic representation alone.}, author = {Sharmanska, Viktoriia and Quadrianto, Novi and Lampert, Christoph}, location = {Florence, Italy}, number = {PART 5}, pages = {242 -- 255}, publisher = {Springer}, title = {{Augmented attribute representations}}, doi = {10.1007/978-3-642-33715-4_18}, volume = {7576}, year = {2012}, } @inproceedings{3126, abstract = {In this work we propose a new information-theoretic clustering algorithm that infers cluster memberships by direct optimization of a non-parametric mutual information estimate between data distribution and cluster assignment. Although the optimization objective has a solid theoretical foundation it is hard to optimize. We propose an approximate optimization formulation that leads to an efficient algorithm with low runtime complexity. The algorithm has a single free parameter, the number of clusters to find. We demonstrate superior performance on several synthetic and real datasets. }, author = {Müller, Andreas and Nowozin, Sebastian and Lampert, Christoph}, location = {Graz, Austria}, pages = {205 -- 215}, publisher = {Springer}, title = {{Information theoretic clustering using minimal spanning trees}}, doi = {10.1007/978-3-642-32717-9_21}, volume = {7476}, year = {2012}, } @article{3248, abstract = {We describe RTblob, a high speed vision system that detects objects in cluttered scenes based on their color and shape at a speed of over 800 frames/s. Because the system is available as open-source software and relies only on off-the-shelf PC hardware components, it can provide the basis for multiple application scenarios. As an illustrative example, we show how RTblob can be used in a robotic table tennis scenario to estimate ball trajectories through 3D space simultaneously from four cameras images at a speed of 200 Hz.}, author = {Lampert, Christoph and Peters, Jan}, issn = {1861-8219}, journal = {Journal of Real-Time Image Processing}, number = {1}, pages = {31 -- 41}, publisher = {Springer}, title = {{Real-time detection of colored objects in multiple camera streams with off-the-shelf hardware components}}, doi = {10.1007/s11554-010-0168-3}, volume = {7}, year = {2012}, } @inproceedings{3124, abstract = {We consider the problem of inference in a graphical model with binary variables. While in theory it is arguably preferable to compute marginal probabilities, in practice researchers often use MAP inference due to the availability of efficient discrete optimization algorithms. We bridge the gap between the two approaches by introducing the Discrete Marginals technique in which approximate marginals are obtained by minimizing an objective function with unary and pairwise terms over a discretized domain. This allows the use of techniques originally developed for MAP-MRF inference and learning. We explore two ways to set up the objective function - by discretizing the Bethe free energy and by learning it from training data. Experimental results show that for certain types of graphs a learned function can outperform the Bethe approximation. We also establish a link between the Bethe free energy and submodular functions. }, author = {Korc, Filip and Kolmogorov, Vladimir and Lampert, Christoph}, location = {Edinburgh, Scotland}, publisher = {ICML}, title = {{Approximating marginals using discrete energy minimization}}, year = {2012}, }