TY - CONF
AB - 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.
AU - Lampert, Christoph
ID - 2825
TI - Dynamic pruning of factor graphs for maximum marginal prediction
VL - 1
ER -
TY - JOUR
AB - We study evolutionary game theory in a setting where individuals learn from each other. We extend the traditional approach by assuming that a population contains individuals with different learning abilities. In particular, we explore the situation where individuals have different search spaces, when attempting to learn the strategies of others. The search space of an individual specifies the set of strategies learnable by that individual. The search space is genetically given and does not change under social evolutionary dynamics. We introduce a general framework and study a specific example in the context of direct reciprocity. For this example, we obtain the counter intuitive result that cooperation can only evolve for intermediate benefit-to-cost ratios, while small and large benefit-to-cost ratios favor defection. Our paper is a step toward making a connection between computational learning theory and evolutionary game dynamics.
AU - Chatterjee, Krishnendu
AU - Zufferey, Damien
AU - Nowak, Martin
ID - 2848
JF - Journal of Theoretical Biology
TI - Evolutionary game dynamics in populations with different learners
VL - 301
ER -
TY - JOUR
AU - Edelsbrunner, Herbert
AU - Strelkova, Nataliya
ID - 2849
IS - 6
JF - Russian Mathematical Surveys
TI - On the configuration space of Steiner minimal trees
VL - 67
ER -
TY - CONF
AB - Formal verification aims to improve the quality of hardware and software by detecting errors before they do harm. At the basis of formal verification lies the logical notion of correctness, which purports to capture whether or not a circuit or program behaves as desired. We suggest that the boolean partition into correct and incorrect systems falls short of the practical need to assess the behavior of hardware and software in a more nuanced fashion against multiple criteria.
AU - Henzinger, Thomas A
ID - 2888
T2 - Conference proceedings MODELS 2012
TI - Quantitative reactive models
VL - 7590
ER -
TY - CONF
AB - Systems are often specified using multiple requirements on their behavior. In practice, these requirements can be contradictory. The classical approach to specification, verification, and synthesis demands more detailed specifications that resolve any contradictions in the requirements. These detailed specifications are usually large, cumbersome, and hard to maintain or modify. In contrast, quantitative frameworks allow the formalization of the intuitive idea that what is desired is an implementation that comes "closest" to satisfying the mutually incompatible requirements, according to a measure of fit that can be defined by the requirements engineer. One flexible framework for quantifying how "well" an implementation satisfies a specification is offered by simulation distances that are parameterized by an error model. We introduce this framework, study its properties, and provide an algorithmic solution for the following quantitative synthesis question: given two (or more) behavioral requirements specified by possibly incompatible finite-state machines, and an error model, find the finite-state implementation that minimizes the maximal simulation distance to the given requirements. Furthermore, we generalize the framework to handle infinite alphabets (for example, realvalued domains). We also demonstrate how quantitative specifications based on simulation distances might lead to smaller and easier to modify specifications. Finally, we illustrate our approach using case studies on error correcting codes and scheduler synthesis.
AU - Cerny, Pavol
AU - Gopi, Sivakanth
AU - Henzinger, Thomas A
AU - Radhakrishna, Arjun
AU - Totla, Nishant
ID - 2890
T2 - Proceedings of the tenth ACM international conference on Embedded software
TI - Synthesis from incompatible specifications
ER -
TY - CONF
AB - Quantitative automata are nondeterministic finite automata with edge weights. They value a
run by some function from the sequence of visited weights to the reals, and value a word by its
minimal/maximal run. They generalize boolean automata, and have gained much attention in
recent years. Unfortunately, important automaton classes, such as sum, discounted-sum, and
limit-average automata, cannot be determinized. Yet, the quantitative setting provides the potential
of approximate determinization. We define approximate determinization with respect to
a distance function, and investigate this potential.
We show that sum automata cannot be determinized approximately with respect to any
distance function. However, restricting to nonnegative weights allows for approximate determinization
with respect to some distance functions.
Discounted-sum automata allow for approximate determinization, as the influence of a word’s
suffix is decaying. However, the naive approach, of unfolding the automaton computations up
to a sufficient level, is shown to be doubly exponential in the discount factor. We provide an
alternative construction that is singly exponential in the discount factor, in the precision, and
in the number of states. We prove matching lower bounds, showing exponential dependency on
each of these three parameters.
Average and limit-average automata are shown to prohibit approximate determinization with
respect to any distance function, and this is the case even for two weights, 0 and 1.
AU - Boker, Udi
AU - Henzinger, Thomas A
ID - 2891
T2 - Leibniz International Proceedings in Informatics
TI - Approximate determinization of quantitative automata
VL - 18
ER -
TY - JOUR
AB - We present an algorithm for simplifying linear cartographic objects and results obtained with a computer program implementing this algorithm.
AU - Edelsbrunner, Herbert
AU - Musin, Oleg
AU - Ukhalov, Alexey
AU - Yakimova, Olga
AU - Alexeev, Vladislav
AU - Bogaevskaya, Victoriya
AU - Gorohov, Andrey
AU - Preobrazhenskaya, Margarita
ID - 2902
IS - 6
JF - Modeling and Analysis of Information Systems
TI - Fractal and computational geometry for generalizing cartographic objects
VL - 19
ER -
TY - CONF
AB - In order to enjoy a digital version of the Jordan Curve Theorem, it is common to use the closed topology for the foreground and the open topology for the background of a 2-dimensional binary image. In this paper, we introduce a single topology that enjoys this theorem for all thresholds decomposing a real-valued image into foreground and background. This topology is easy to construct and it generalizes to n-dimensional images.
AU - Edelsbrunner, Herbert
AU - Symonova, Olga
ID - 2903
TI - The adaptive topology of a digital image
ER -
TY - JOUR
AB - Generalized van der Corput sequences are onedimensional, infinite sequences in the unit interval. They are generated from permutations in integer base b and are the building blocks of the multi-dimensional Halton sequences. Motivated by recent progress of Atanassov on the uniform distribution behavior of Halton sequences, we study, among others, permutations of the form P(i) = ai (mod b) for coprime integers a and b. We show that multipliers a that either divide b - 1 or b + 1 generate van der Corput sequences with weak distribution properties. We give explicit lower bounds for the asymptotic distribution behavior of these sequences and relate them to sequences generated from the identity permutation in smaller bases, which are, due to Faure, the weakest distributed generalized van der Corput sequences.
AU - Pausinger, Florian
ID - 2904
IS - 3
JF - Journal de Theorie des Nombres des Bordeaux
SN - 2118-8572
TI - Weak multipliers for generalized van der Corput sequences
VL - 24
ER -
TY - JOUR
AU - Edelsbrunner, Herbert
AU - Strelkova, Nataliya
ID - 2912
IS - 6
JF - Uspekhi Mat. Nauk
TI - Configuration space for shortest networks
VL - 67
ER -
TY - CONF
AU - Kroemer, Oliver
AU - Lampert, Christoph
AU - Peters, Jan
ID - 2915
TI - Multi-modal learning for dynamic tactile sensing
ER -
TY - CONF
AB - The classical (boolean) notion of refinement for behavioral interfaces of system components is the alternating refinement preorder. In this paper, we define a quantitative measure for interfaces, called interface simulation distance. It makes the alternating refinement preorder quantitative by, intu- itively, tolerating errors (while counting them) in the alternating simulation game. We show that the interface simulation distance satisfies the triangle inequality, that the distance between two interfaces does not increase under parallel composition with a third interface, and that the distance between two interfaces can be bounded from above and below by distances between abstractions of the two interfaces. We illustrate the framework, and the properties of the distances under composition of interfaces, with two case studies.
AU - Cerny, Pavol
AU - Chmelik, Martin
AU - Henzinger, Thomas A
AU - Radhakrishna, Arjun
ID - 2916
T2 - Electronic Proceedings in Theoretical Computer Science
TI - Interface Simulation Distances
VL - 96
ER -
TY - JOUR
AB - The search for extra-terrestrial intelligence (SETI) has been performed principally as a one-way survey, listening of radio frequencies across the Milky Way and other galaxies. However, scientists have engaged in an active messaging only rarely. This suggests the simple rationale that if other civilizations exist and take a similar approach to ours, namely listening but not broadcasting, the result is a silent universe. A simple game theoretical model, the prisoner's dilemma, explains this situation: each player (civilization) can passively search (defect), or actively search and broadcast (cooperate). In order to maximize the payoff (or, equivalently, minimize the risks) the best strategy is not to broadcast. In fact, the active search has been opposed on the basis that it might be dangerous to expose ourselves. However, most of these ideas have not been based on objective arguments, and ignore accounting of the possible gains and losses. Thus, the question stands: should we perform an active search? I develop a game-theoretical framework where civilizations can be of different types, and explicitly apply it to a situation where societies are either interested in establishing a two-way communication or belligerent and in urge to exploit ours. The framework gives a quantitative solution (a mixed-strategy), which is how frequent we should perform the active SETI. This frequency is roughly proportional to the inverse of the risk, and can be extremely small. However, given the immense amount of stars being scanned, it supports active SETI. The model is compared with simulations, and the possible actions are evaluated through the San Marino scale, measuring the risks of messaging.
AU - Vladar, Harold
ID - 2917
IS - 1
JF - International Journal of Astrobiology
TI - The game of active search for extra terrestrial intelligence Breaking the Great Silence
VL - 12
ER -
TY - GEN
AB - This paper addresses the problem of approximate MAP-MRF inference in general graphical models. Following [36], we consider a family of linear programming relaxations of the problem where each relaxation is specified by a set of nested pairs of factors for which the marginalization constraint needs to be enforced. We develop a generalization of the TRW-S algorithm [9] for this problem, where we use a decomposition into junction chains, monotonic w.r.t. some ordering on the nodes. This generalizes the monotonic chains in [9] in a natural way. We also show how to deal with nested factors in an efficient way. Experiments show an improvement over min-sum diffusion, MPLP and subgradient ascent algorithms on a number of computer vision and natural language processing problems.
AU - Kolmogorov, Vladimir
AU - Schoenemann, Thomas
ID - 2928
T2 - arXiv
TI - Generalized sequential tree-reweighted message passing
ER -
TY - CONF
AB - In this paper we investigate k-submodular functions. This natural family of discrete functions includes submodular and bisubmodular functions as the special cases k = 1 and k = 2 respectively.
In particular we generalize the known Min-Max-Theorem for submodular and bisubmodular functions. This theorem asserts that the minimum of the (bi)submodular function can be found by solving a maximization problem over a (bi)submodular polyhedron. We define a k-submodular polyhedron, prove a Min-Max-Theorem for k-submodular functions, and give a greedy algorithm to construct the vertices of the polyhedron.
AU - Huber, Anna
AU - Kolmogorov, Vladimir
ID - 2930
TI - Towards minimizing k-submodular functions
VL - 7422
ER -
TY - JOUR
AB - In this paper, we present a new approach for establishing correspondences between sparse image features related by an unknown nonrigid mapping and corrupted by clutter and occlusion, such as points extracted from images of different instances of the same object category. We formulate this matching task as an energy minimization problem by defining an elaborate objective function of the appearance and the spatial arrangement of the features. Optimization of this energy is an instance of graph matching, which is in general an NP-hard problem. We describe a novel graph matching optimization technique, which we refer to as dual decomposition (DD), and demonstrate on a variety of examples that this method outperforms existing graph matching algorithms. In the majority of our examples, DD is able to find the global minimum within a minute. The ability to globally optimize the objective allows us to accurately learn the parameters of our matching model from training examples. We show on several matching tasks that our learned model yields results superior to those of state-of-the-art methods.
AU - Torresani, Lorenzo
AU - Kolmogorov, Vladimir
AU - Rother, Carsten
ID - 2931
IS - 2
JF - IEEE Transactions on Pattern Analysis and Machine Intelligence
TI - A dual decomposition approach to feature correspondence
VL - 35
ER -
TY - CONF
AB - The notion of delays arises naturally in many computational models, such as, in the design of circuits, control systems, and dataflow languages. In this work, we introduce automata with delay blocks (ADBs), extending finite state automata with variable time delay blocks, for deferring individual transition output symbols, in a discrete-time setting. We show that the ADB languages strictly subsume the regular languages, and are incomparable in expressive power to the context-free languages. We show that ADBs are closed under union, concatenation and Kleene star, and under intersection with regular languages, but not closed under complementation and intersection with other ADB languages. We show that the emptiness and the membership problems are decidable in polynomial time for ADBs, whereas the universality problem is undecidable. Finally we consider the linear-time model checking problem, i.e., whether the language of an ADB is contained in a regular language, and show that the model checking problem is PSPACE-complete. Copyright 2012 ACM.
AU - Chatterjee, Krishnendu
AU - Henzinger, Thomas A
AU - Prabhu, Vinayak
ID - 2936
T2 - roceedings of the tenth ACM international conference on Embedded software
TI - Finite automata with time delay blocks
ER -
TY - CONF
AB - Developers building cryptography into security-sensitive applications face a daunting task. Not only must they understand the security guarantees delivered by the constructions they choose, they must also implement and combine them correctly and efficiently. Cryptographic compilers free developers from this task by turning high-level specifications of security goals into efficient implementations. Yet, trusting such tools is hard as they rely on complex mathematical machinery and claim security properties that are subtle and difficult to verify. In this paper we present ZKCrypt, an optimizing cryptographic compiler achieving an unprecedented level of assurance without sacrificing practicality for a comprehensive class of cryptographic protocols, known as Zero-Knowledge Proofs of Knowledge. The pipeline of ZKCrypt integrates purpose-built verified compilers and verifying compilers producing formal proofs in the CertiCrypt framework. By combining the guarantees delivered by each stage, ZKCrypt provides assurance that the output implementation securely realizes the abstract proof goal given as input. We report on the main characteristics of ZKCrypt, highlight new definitions and concepts at its foundations, and illustrate its applicability through a representative example of an anonymous credential system.
AU - Almeida, José
AU - Barbosa, Manuel
AU - Bangerter, Endre
AU - Barthe, Gilles
AU - Krenn, Stephan
AU - Béguelin, Santiago
ID - 2937
T2 - Proceedings of the 2012 ACM conference on Computer and communications security
TI - Full proof cryptography: Verifiable compilation of efficient zero-knowledge protocols
ER -
TY - JOUR
AB - Social insects have a very high potential to become invasive pest species. Here, we explore how their social lifestyle and their interaction with parasites may contribute to this invasive success. Similar to solitary species, parasite release followed by the evolution of increased competitive ability can promote establishment of introduced social insect hosts in their introduced range. Genetic bottlenecks during introduction of low numbers of founder individuals decrease the genetic diversity at three levels: the population, the colony and the individual, with the colony level being specific to social insects. Reduced genetic diversity can affect both the individual immune system and the collective colony-level disease defences (social immunity). Still, the dual immune system is likely to make social insects more robust to parasite attack. Changes in social structure from small, family-based, territorially aggressive societies in native populations towards huge networks of cooperating nests (unicoloniality) occur in some invasive social insects, for example, most invasive ants and some termites. Unicoloniality is likely to affect disease dynamics in multiple ways. The free exchange of individuals within the population leads to an increased genetic heterogeneity among individuals of a single nest, thereby decreasing disease transmission. However, the multitude of reproductively active queens per colony buffers the effect of individual diseased queens and their offspring, which may result in a higher level of vertical disease transmission in unicolonial societies. Lastly, unicoloniality provides a competitive advantage over native species, allowing them to quickly become the dominant species in the habitat, which in turn selects for parasite adaptation to this common host genotype and thus eventually a high parasite pressure. Overall, invasions by insect societies are characterized by general features applying to all introduced species, as well as idiosyncrasies that emerge from their social lifestyle. It is important to study these effects in concert to be able to develop efficient management and biocontrol strategies. © 2012 British Ecological Society.
AU - Ugelvig, Line V
AU - Cremer, Sylvia
ID - 2938
IS - 6
JF - Functional Ecology
TI - Effects of social immunity and unicoloniality on host parasite interactions in invasive insect societies
VL - 26
ER -
TY - JOUR
AU - Dolbilin, Nikolai
AU - Edelsbrunner, Herbert
AU - Musin, Oleg
ID - 2941
IS - 4
JF - Russian Mathematical Surveys
TI - On the optimality of functionals over triangulations of Delaunay sets
VL - 67
ER -