TY - CONF AB - 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. AU - Kolesnikov, Alexander AU - Kuznetsova, Alina AU - Lampert, Christoph AU - Ferrari, Vittorio ID - 7640 SN - 9781728150239 T2 - Proceedings of the 2019 International Conference on Computer Vision Workshop TI - Detecting visual relationships using box attention ER - TY - CONF AB - Deep neural networks (DNNs) have become increasingly important due to their excellent empirical performance on a wide range of problems. However, regularization is generally achieved by indirect means, largely due to the complex set of functions defined by a network and the difficulty in measuring function complexity. There exists no method in the literature for additive regularization based on a norm of the function, as is classically considered in statistical learning theory. In this work, we study the tractability of function norms for deep neural networks with ReLU activations. We provide, to the best of our knowledge, the first proof in the literature of the NP-hardness of computing function norms of DNNs of 3 or more layers. We also highlight a fundamental difference between shallow and deep networks. In the light on these results, we propose a new regularization strategy based on approximate function norms, and show its efficiency on a segmentation task with a DNN. AU - Rannen-Triki, Amal AU - Berman, Maxim AU - Kolmogorov, Vladimir AU - Blaschko, Matthew B. ID - 7639 SN - 9781728150239 T2 - Proceedings of the 2019 International Conference on Computer Vision Workshop TI - Function norms for neural networks ER - TY - CHAP AB - We review the history of population genetics, starting with its origins a century ago from the synthesis between Mendel and Darwin's ideas, through to the recent development of sophisticated schemes of inference from sequence data, based on the coalescent. We explain the close relation between the coalescent and a diffusion process, which we illustrate by their application to understand spatial structure. We summarise the powerful methods available for analysis of multiple loci, when linkage equilibrium can be assumed, and then discuss approaches to the more challenging case, where associations between alleles require that we follow genotype, rather than allele, frequencies. Though we can hardly cover the whole of population genetics, we give an overview of the current state of the subject, and future challenges to it. AU - Barton, Nicholas H AU - Etheridge, Alison ED - Balding, David ED - Moltke, Ida ED - Marioni, John ID - 8281 SN - 9781119429142 T2 - Handbook of statistical genomics TI - Mathematical models in population genetics ER - TY - GEN AB - Denote by ∆N the N-dimensional simplex. A map f : ∆N → Rd is an almost r-embedding if fσ1∩. . .∩fσr = ∅ whenever σ1, . . . , σr are pairwise disjoint faces. A counterexample to the topological Tverberg conjecture asserts that if r is not a prime power and d ≥ 2r + 1, then there is an almost r-embedding ∆(d+1)(r−1) → Rd. This was improved by Blagojevi´c–Frick–Ziegler using a simple construction of higher-dimensional counterexamples by taking k-fold join power of lower-dimensional ones. We improve this further (for d large compared to r): If r is not a prime power and N := (d+ 1)r−r l d + 2 r + 1 m−2, then there is an almost r-embedding ∆N → Rd. For the r-fold van Kampen–Flores conjecture we also produce counterexamples which are stronger than previously known. Our proof is based on generalizations of the Mabillard–Wagner theorem on construction of almost r-embeddings from equivariant maps, and of the Ozaydin theorem on existence of equivariant maps. AU - Avvakumov, Sergey AU - Karasev, R. AU - Skopenkov, A. ID - 8184 T2 - arXiv TI - Stronger counterexamples to the topological Tverberg conjecture ER - TY - CONF AB - A proxy re-encryption (PRE) scheme is a public-key encryption scheme that allows the holder of a key pk to derive a re-encryption key for any other key 𝑝𝑘′. This re-encryption key lets anyone transform ciphertexts under pk into ciphertexts under 𝑝𝑘′ without having to know the underlying message, while transformations from 𝑝𝑘′ to pk should not be possible (unidirectional). Security is defined in a multi-user setting against an adversary that gets the users’ public keys and can ask for re-encryption keys and can corrupt users by requesting their secret keys. Any ciphertext that the adversary cannot trivially decrypt given the obtained secret and re-encryption keys should be secure. All existing security proofs for PRE only show selective security, where the adversary must first declare the users it wants to corrupt. This can be lifted to more meaningful adaptive security by guessing the set of corrupted users among the n users, which loses a factor exponential in Open image in new window , rendering the result meaningless already for moderate Open image in new window . Jafargholi et al. (CRYPTO’17) proposed a framework that in some cases allows to give adaptive security proofs for schemes which were previously only known to be selectively secure, while avoiding the exponential loss that results from guessing the adaptive choices made by an adversary. We apply their framework to PREs that satisfy some natural additional properties. Concretely, we give a more fine-grained reduction for several unidirectional PREs, proving adaptive security at a much smaller loss. The loss depends on the graph of users whose edges represent the re-encryption keys queried by the adversary. For trees and chains the loss is quasi-polynomial in the size and for general graphs it is exponential in their depth and indegree (instead of their size as for previous reductions). Fortunately, trees and low-depth graphs cover many, if not most, interesting applications. Our results apply e.g. to the bilinear-map based PRE schemes by Ateniese et al. (NDSS’05 and CT-RSA’09), Gentry’s FHE-based scheme (STOC’09) and the LWE-based scheme by Chandran et al. (PKC’14). AU - Fuchsbauer, Georg AU - Kamath Hosdurg, Chethan AU - Klein, Karen AU - Pietrzak, Krzysztof Z ID - 6430 SN - 03029743 TI - Adaptively secure proxy re-encryption VL - 11443 ER -