TY - CONF AB - 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. AU - Royer, Amélie AU - Lampert, Christoph ID - 7936 SN - 9781728165530 T2 - IEEE Winter Conference on Applications of Computer Vision TI - Localizing grouped instances for efficient detection in low-resource scenarios ER - TY - CONF AB - 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. AU - Royer, Amélie AU - Lampert, Christoph ID - 7937 SN - 9781728165530 T2 - 2020 IEEE Winter Conference on Applications of Computer Vision TI - A flexible selection scheme for minimum-effort transfer learning ER - TY - CONF AB - Multiple-environment Markov decision processes (MEMDPs) are MDPs equipped with not one, but multiple probabilistic transition functions, which represent the various possible unknown environments. While the previous research on MEMDPs focused on theoretical properties for long-run average payoff, we study them with discounted-sum payoff and focus on their practical advantages and applications. MEMDPs can be viewed as a special case of Partially observable and Mixed observability MDPs: the state of the system is perfectly observable, but not the environment. We show that the specific structure of MEMDPs allows for more efficient algorithmic analysis, in particular for faster belief updates. We demonstrate the applicability of MEMDPs in several domains. In particular, we formalize the sequential decision-making approach to contextual recommendation systems as MEMDPs and substantially improve over the previous MDP approach. AU - Chatterjee, Krishnendu AU - Chmelik, Martin AU - Karkhanis, Deep AU - Novotný, Petr AU - Royer, Amélie ID - 8193 SN - 23340835 T2 - Proceedings of the 30th International Conference on Automated Planning and Scheduling TI - Multiple-environment Markov decision processes: Efficient analysis and applications VL - 30 ER - TY - CHAP AB - 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. AU - Royer, Amélie AU - Bousmalis, Konstantinos AU - Gouws, Stephan AU - Bertsch, Fred AU - Mosseri, Inbar AU - Cole, Forrester AU - Murphy, Kevin ED - Singh, Richa ED - Vatsa, Mayank ED - Patel, Vishal M. ED - Ratha, Nalini ID - 8092 SN - 9783030306717 T2 - Domain Adaptation for Visual Understanding TI - XGAN: Unsupervised image-to-image translation for many-to-many mappings ER - TY - THES AB - This thesis considers two examples of reconfiguration problems: flipping edges in edge-labelled triangulations of planar point sets and swapping labelled tokens placed on vertices of a graph. In both cases the studied structures – all the triangulations of a given point set or all token placements on a given graph – can be thought of as vertices of the so-called reconfiguration graph, in which two vertices are adjacent if the corresponding structures differ by a single elementary operation – by a flip of a diagonal in a triangulation or by a swap of tokens on adjacent vertices, respectively. We study the reconfiguration of one instance of a structure into another via (shortest) paths in the reconfiguration graph. For triangulations of point sets in which each edge has a unique label and a flip transfers the label from the removed edge to the new edge, we prove a polynomial-time testable condition, called the Orbit Theorem, that characterizes when two triangulations of the same point set lie in the same connected component of the reconfiguration graph. The condition was first conjectured by Bose, Lubiw, Pathak and Verdonschot. We additionally provide a polynomial time algorithm that computes a reconfiguring flip sequence, if it exists. Our proof of the Orbit Theorem uses topological properties of a certain high-dimensional cell complex that has the usual reconfiguration graph as its 1-skeleton. In the context of token swapping on a tree graph, we make partial progress on the problem of finding shortest reconfiguration sequences. We disprove the so-called Happy Leaf Conjecture and demonstrate the importance of swapping tokens that are already placed at the correct vertices. We also prove that a generalization of the problem to weighted coloured token swapping is NP-hard on trees but solvable in polynomial time on paths and stars. AU - Masárová, Zuzana ID - 7944 KW - reconfiguration KW - reconfiguration graph KW - triangulations KW - flip KW - constrained triangulations KW - shellability KW - piecewise-linear balls KW - token swapping KW - trees KW - coloured weighted token swapping SN - 2663-337X TI - Reconfiguration problems ER - TY - JOUR AB - Inspired by the possibility to experimentally manipulate and enhance chemical reactivity in helium nanodroplets, we investigate the effective interaction and the resulting correlations between two diatomic molecules immersed in a bath of bosons. By analogy with the bipolaron, we introduce the biangulon quasiparticle describing two rotating molecules that align with respect to each other due to the effective attractive interaction mediated by the excitations of the bath. We study this system in different parameter regimes and apply several theoretical approaches to describe its properties. Using a Born–Oppenheimer approximation, we investigate the dependence of the effective intermolecular interaction on the rotational state of the two molecules. In the strong-coupling regime, a product-state ansatz shows that the molecules tend to have a strong alignment in the ground state. To investigate the system in the weak-coupling regime, we apply a one-phonon excitation variational ansatz, which allows us to access the energy spectrum. In comparison to the angulon quasiparticle, the biangulon shows shifted angulon instabilities and an additional spectral instability, where resonant angular momentum transfer between the molecules and the bath takes place. These features are proposed as an experimentally observable signature for the formation of the biangulon quasiparticle. Finally, by using products of single angulon and bare impurity wave functions as basis states, we introduce a diagonalization scheme that allows us to describe the transition from two separated angulons to a biangulon as a function of the distance between the two molecules. AU - Li, Xiang AU - Yakaboylu, Enderalp AU - Bighin, Giacomo AU - Schmidt, Richard AU - Lemeshko, Mikhail AU - Deuchert, Andreas ID - 8587 IS - 16 JF - The Journal of Chemical Physics KW - Physical and Theoretical Chemistry KW - General Physics and Astronomy SN - 0021-9606 TI - Intermolecular forces and correlations mediated by a phonon bath VL - 152 ER - TY - THES AB - One of the most striking hallmarks of the eukaryotic cell is the presence of intracellular vesicles and organelles. Each of these membrane-enclosed compartments has a distinct composition of lipids and proteins, which is essential for accurate membrane traffic and homeostasis. Interestingly, their biochemical identities are achieved with the help of small GTPases of the Rab family, which cycle between GDP- and GTP-bound forms on the selected membrane surface. While this activity switch is well understood for an individual protein, how Rab GTPases collectively transition between states to generate decisive signal propagation in space and time is unclear. In my PhD thesis, I present in vitro reconstitution experiments with theoretical modeling to systematically study a minimal Rab5 activation network from bottom-up. We find that positive feedback based on known molecular interactions gives rise to bistable GTPase activity switching on system’s scale. Furthermore, we determine that collective transition near the critical point is intrinsically stochastic and provide evidence that the inactive Rab5 abundance on the membrane can shape the network response. Finally, we demonstrate that collective switching can spread on the lipid bilayer as a traveling activation wave, representing a possible emergent activity pattern in endosomal maturation. Together, our findings reveal new insights into the self-organization properties of signaling networks away from chemical equilibrium. Our work highlights the importance of systematic characterization of biochemical systems in well-defined physiological conditions. This way, we were able to answer long-standing open questions in the field and close the gap between regulatory processes on a molecular scale and emergent responses on system’s level. AU - Bezeljak, Urban ID - 8341 SN - 2663-337X TI - In vitro reconstitution of a Rab activation switch ER - TY - JOUR AB - The eukaryotic endomembrane system is controlled by small GTPases of the Rab family, which are activated at defined times and locations in a switch-like manner. While this switch is well understood for an individual protein, how regulatory networks produce intracellular activity patterns is currently not known. Here, we combine in vitro reconstitution experiments with computational modeling to study a minimal Rab5 activation network. We find that the molecular interactions in this system give rise to a positive feedback and bistable collective switching of Rab5. Furthermore, we find that switching near the critical point is intrinsically stochastic and provide evidence that controlling the inactive population of Rab5 on the membrane can shape the network response. Notably, we demonstrate that collective switching can spread on the membrane surface as a traveling wave of Rab5 activation. Together, our findings reveal how biochemical signaling networks control vesicle trafficking pathways and how their nonequilibrium properties define the spatiotemporal organization of the cell. AU - Bezeljak, Urban AU - Loya, Hrushikesh AU - Kaczmarek, Beata M AU - Saunders, Timothy E. AU - Loose, Martin ID - 7580 IS - 12 JF - Proceedings of the National Academy of Sciences SN - 0027-8424 TI - Stochastic activation and bistability in a Rab GTPase regulatory network VL - 117 ER - TY - THES AB - Algorithms in computational 3-manifold topology typically take a triangulation as an input and return topological information about the underlying 3-manifold. However, extracting the desired information from a triangulation (e.g., evaluating an invariant) is often computationally very expensive. In recent years this complexity barrier has been successfully tackled in some cases by importing ideas from the theory of parameterized algorithms into the realm of 3-manifolds. Various computationally hard problems were shown to be efficiently solvable for input triangulations that are sufficiently “tree-like.” In this thesis we focus on the key combinatorial parameter in the above context: we consider the treewidth of a compact, orientable 3-manifold, i.e., the smallest treewidth of the dual graph of any triangulation thereof. By building on the work of Scharlemann–Thompson and Scharlemann–Schultens–Saito on generalized Heegaard splittings, and on the work of Jaco–Rubinstein on layered triangulations, we establish quantitative relations between the treewidth and classical topological invariants of a 3-manifold. In particular, among other results, we show that the treewidth of a closed, orientable, irreducible, non-Haken 3-manifold is always within a constant factor of its Heegaard genus. AU - Huszár, Kristóf ID - 8032 SN - 2663-337X TI - Combinatorial width parameters for 3-dimensional manifolds ER - TY - CONF AB - This paper presents a foundation for refining concurrent programs with structured control flow. The verification problem is decomposed into subproblems that aid interactive program development, proof reuse, and automation. The formalization in this paper is the basis of a new design and implementation of the Civl verifier. AU - Kragl, Bernhard AU - Qadeer, Shaz AU - Henzinger, Thomas A ID - 8195 SN - 0302-9743 T2 - Computer Aided Verification TI - Refinement for structured concurrent programs VL - 12224 ER - TY - CONF AB - Asynchronous programs are notoriously difficult to reason about because they spawn computation tasks which take effect asynchronously in a nondeterministic way. Devising inductive invariants for such programs requires understanding and stating complex relationships between an unbounded number of computation tasks in arbitrarily long executions. In this paper, we introduce inductive sequentialization, a new proof rule that sidesteps this complexity via a sequential reduction, a sequential program that captures every behavior of the original program up to reordering of coarse-grained commutative actions. A sequential reduction of a concurrent program is easy to reason about since it corresponds to a simple execution of the program in an idealized synchronous environment, where processes act in a fixed order and at the same speed. We have implemented and integrated our proof rule in the CIVL verifier, allowing us to provably derive fine-grained implementations of asynchronous programs. We have successfully applied our proof rule to a diverse set of message-passing protocols, including leader election protocols, two-phase commit, and Paxos. AU - Kragl, Bernhard AU - Enea, Constantin AU - Henzinger, Thomas A AU - Mutluergil, Suha Orhun AU - Qadeer, Shaz ID - 8012 SN - 9781450376136 T2 - Proceedings of the 41st ACM SIGPLAN Conference on Programming Language Design and Implementation TI - Inductive sequentialization of asynchronous programs ER - TY - THES AB - During bacterial cell division, the tubulin-homolog FtsZ forms a ring-like structure at the center of the cell. This so-called Z-ring acts as a scaffold recruiting several division-related proteins to mid-cell and plays a key role in distributing proteins at the division site, a feature driven by the treadmilling motion of FtsZ filaments around the septum. What regulates the architecture, dynamics and stability of the Z-ring is still poorly understood, but FtsZ-associated proteins (Zaps) are known to play an important role. Advances in fluorescence microscopy and in vitro reconstitution experiments have helped to shed light into some of the dynamic properties of these complex systems, but methods that allow to collect and analyze large quantitative data sets of the underlying polymer dynamics are still missing. Here, using an in vitro reconstitution approach, we studied how different Zaps affect FtsZ filament dynamics and organization into large-scale patterns, giving special emphasis to the role of the well-conserved protein ZapA. For this purpose, we use high-resolution fluorescence microscopy combined with novel image analysis workfows to study pattern organization and polymerization dynamics of active filaments. We quantified the influence of Zaps on FtsZ on three diferent spatial scales: the large-scale organization of the membrane-bound filament network, the underlying polymerization dynamics and the behavior of single molecules. We found that ZapA cooperatively increases the spatial order of the filament network, binds only transiently to FtsZ filaments and has no effect on filament length and treadmilling velocity. Our data provides a model for how FtsZ-associated proteins can increase the precision and stability of the bacterial cell division machinery in a switch-like manner, without compromising filament dynamics. Furthermore, we believe that our automated quantitative methods can be used to analyze a large variety of dynamic cytoskeletal systems, using standard time-lapse movies of homogeneously labeled proteins obtained from experiments in vitro or even inside the living cell. AU - Dos Santos Caldas, Paulo R ID - 8358 SN - 2663-337X TI - Organization and dynamics of treadmilling filaments in cytoskeletal networks of FtsZ and its crosslinkers ER - TY - CONF AB - Even though Delaunay originally introduced his famous triangulations in the case of infinite point sets with translational periodicity, a software that computes such triangulations in the general case is not yet available, to the best of our knowledge. Combining and generalizing previous work, we present a practical algorithm for computing such triangulations. The algorithm has been implemented and experiments show that its performance is as good as the one of the CGAL package, which is restricted to cubic periodicity. AU - Osang, Georg F AU - Rouxel-Labbé, Mael AU - Teillaud, Monique ID - 8703 SN - 18688969 T2 - 28th Annual European Symposium on Algorithms TI - Generalizing CGAL periodic Delaunay triangulations VL - 173 ER - TY - CONF AB - 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. AU - Bui Thi Mai, Phuong AU - Lampert, Christoph ID - 7481 T2 - 8th International Conference on Learning Representations TI - Functional vs. parametric equivalence of ReLU networks ER - TY - JOUR AB - We consider the Pekar functional on a ball in ℝ3. We prove uniqueness of minimizers, and a quadratic lower bound in terms of the distance to the minimizer. The latter follows from nondegeneracy of the Hessian at the minimum. AU - Feliciangeli, Dario AU - Seiringer, Robert ID - 9781 IS - 1 JF - SIAM Journal on Mathematical Analysis KW - Applied Mathematics KW - Computational Mathematics KW - Analysis SN - 0036-1410 TI - Uniqueness and nondegeneracy of minimizers of the Pekar functional on a ball VL - 52 ER - TY - JOUR AB - In the present work, we consider the evolution of two fluids separated by a sharp interface in the presence of surface tension—like, for example, the evolution of oil bubbles in water. Our main result is a weak–strong uniqueness principle for the corresponding free boundary problem for the incompressible Navier–Stokes equation: as long as a strong solution exists, any varifold solution must coincide with it. In particular, in the absence of physical singularities, the concept of varifold solutions—whose global in time existence has been shown by Abels (Interfaces Free Bound 9(1):31–65, 2007) for general initial data—does not introduce a mechanism for non-uniqueness. The key ingredient of our approach is the construction of a relative entropy functional capable of controlling the interface error. If the viscosities of the two fluids do not coincide, even for classical (strong) solutions the gradient of the velocity field becomes discontinuous at the interface, introducing the need for a careful additional adaption of the relative entropy. AU - Fischer, Julian L AU - Hensel, Sebastian ID - 7489 JF - Archive for Rational Mechanics and Analysis SN - 00039527 TI - Weak–strong uniqueness for the Navier–Stokes equation for two fluids with surface tension VL - 236 ER - TY - GEN AB - We prove that in the absence of topological changes, the notion of BV solutions to planar multiphase mean curvature flow does not allow for a mechanism for (unphysical) non-uniqueness. Our approach is based on the local structure of the energy landscape near a classical evolution by mean curvature. Mean curvature flow being the gradient flow of the surface energy functional, we develop a gradient-flow analogue of the notion of calibrations. Just like the existence of a calibration guarantees that one has reached a global minimum in the energy landscape, the existence of a "gradient flow calibration" ensures that the route of steepest descent in the energy landscape is unique and stable. AU - Fischer, Julian L AU - Hensel, Sebastian AU - Laux, Tim AU - Simon, Thilo ID - 10012 T2 - arXiv TI - The local structure of the energy landscape in multiphase mean curvature flow: weak-strong uniqueness and stability of evolutions ER - TY - JOUR AB - The superconducting circuit community has recently discovered the promising potential of superinductors. These circuit elements have a characteristic impedance exceeding the resistance quantum RQ ≈ 6.45 kΩ which leads to a suppression of ground state charge fluctuations. Applications include the realization of hardware protected qubits for fault tolerant quantum computing, improved coupling to small dipole moment objects and defining a new quantum metrology standard for the ampere. In this work we refute the widespread notion that superinductors can only be implemented based on kinetic inductance, i.e. using disordered superconductors or Josephson junction arrays. We present modeling, fabrication and characterization of 104 planar aluminum coil resonators with a characteristic impedance up to 30.9 kΩ at 5.6 GHz and a capacitance down to ≤ 1 fF, with lowloss and a power handling reaching 108 intra-cavity photons. Geometric superinductors are free of uncontrolled tunneling events and offer high reproducibility, linearity and the ability to couple magnetically - properties that significantly broaden the scope of future quantum circuits. AU - Peruzzo, Matilda AU - Trioni, Andrea AU - Hassani, Farid AU - Zemlicka, Martin AU - Fink, Johannes M ID - 8755 IS - 4 JF - Physical Review Applied TI - Surpassing the resistance quantum with a geometric superinductor VL - 14 ER - TY - JOUR AB - This paper deals with dynamical optimal transport metrics defined by spatial discretisation of the Benamou–Benamou formula for the Kantorovich metric . Such metrics appear naturally in discretisations of -gradient flow formulations for dissipative PDE. However, it has recently been shown that these metrics do not in general converge to , unless strong geometric constraints are imposed on the discrete mesh. In this paper we prove that, in a 1-dimensional periodic setting, discrete transport metrics converge to a limiting transport metric with a non-trivial effective mobility. This mobility depends sensitively on the geometry of the mesh and on the non-local mobility at the discrete level. Our result quantifies to what extent discrete transport can make use of microstructure in the mesh to reduce the cost of transport. AU - Gladbach, Peter AU - Kopfer, Eva AU - Maas, Jan AU - Portinale, Lorenzo ID - 7573 IS - 7 JF - Journal de Mathematiques Pures et Appliquees SN - 00217824 TI - Homogenisation of one-dimensional discrete optimal transport VL - 139 ER - TY - GEN AB - We consider finite-volume approximations of Fokker-Planck equations on bounded convex domains in R^d and study the corresponding gradient flow structures. We reprove the convergence of the discrete to continuous Fokker-Planck equation via the method of Evolutionary Γ-convergence, i.e., we pass to the limit at the level of the gradient flow structures, generalising the one-dimensional result obtained by Disser and Liero. The proof is of variational nature and relies on a Mosco convergence result for functionals in the discrete-to-continuum limit that is of independent interest. Our results apply to arbitrary regular meshes, even though the associated discrete transport distances may fail to converge to the Wasserstein distance in this generality. AU - Forkert, Dominik L AU - Maas, Jan AU - Portinale, Lorenzo ID - 10022 T2 - arXiv TI - Evolutionary Γ-convergence of entropic gradient flow structures for Fokker-Planck equations in multiple dimensions ER -