TY - JOUR
AB - Powerful statistical models that can be learned efficiently from large amounts of data are currently revolutionizing computer vision. These models possess a rich internal structure reflecting task-specific relations and constraints. This monograph introduces the reader to the most popular classes of structured models in computer vision. Our focus is discrete undirected graphical models which we cover in detail together with a description of algorithms for both probabilistic inference and maximum a posteriori inference. We discuss separately recently successful techniques for prediction in general structured models. In the second part of this monograph we describe methods for parameter learning where we distinguish the classic maximum likelihood based methods from the more recent prediction-based parameter learning methods. We highlight developments to enhance current models and discuss kernelized models and latent variable models. To make the monograph more practical and to provide links to further study we provide examples of successful application of many methods in the computer vision literature.
AU - Nowozin, Sebastian
AU - Lampert, Christoph
ID - 3320
IS - 3-4
JF - Foundations and Trends in Computer Graphics and Vision
TI - Structured learning and prediction in computer vision
VL - 6
ER -
TY - CONF
AB - Automated termination provers often use the following schema to prove that a program terminates: construct a relational abstraction of the program's transition relation and then show that the relational abstraction is well-founded. The focus of current tools has been on developing sophisticated techniques for constructing the abstractions while relying on known decidable logics (such as linear arithmetic) to express them. We believe we can significantly increase the class of programs that are amenable to automated termination proofs by identifying more expressive decidable logics for reasoning about well-founded relations. We therefore present a new decision procedure for reasoning about multiset orderings, which are among the most powerful orderings used to prove termination. We show that, using our decision procedure, one can automatically prove termination of natural abstractions of programs.
AU - Piskac, Ruzica
AU - Wies, Thomas
ED - Jhala, Ranjit
ED - Schmidt, David
ID - 3324
TI - Decision procedures for automating termination proofs
VL - 6538
ER -
TY - CONF
AB - We introduce streaming data string transducers that map input data strings to output data strings in a single left-to-right pass in linear time. Data strings are (unbounded) sequences of data values, tagged with symbols from a finite set, over a potentially infinite data do- main that supports only the operations of equality and ordering. The transducer uses a finite set of states, a finite set of variables ranging over the data domain, and a finite set of variables ranging over data strings. At every step, it can make decisions based on the next in- put symbol, updating its state, remembering the input data value in its data variables, and updating data-string variables by concatenat- ing data-string variables and new symbols formed from data vari- ables, while avoiding duplication. We establish that the problems of checking functional equivalence of two streaming transducers, and of checking whether a streaming transducer satisfies pre/post verification conditions specified by streaming acceptors over in- put/output data-strings, are in PSPACE. We identify a class of imperative and a class of functional pro- grams, manipulating lists of data items, which can be effectively translated to streaming data-string transducers. The imperative pro- grams dynamically modify a singly-linked heap by changing next- pointers of heap-nodes and by adding new nodes. The main re- striction specifies how the next-pointers can be used for traversal. We also identify an expressively equivalent fragment of functional programs that traverse a list using syntactically restricted recursive calls. Our results lead to algorithms for assertion checking and for checking functional equivalence of two programs, written possibly in different programming styles, for commonly used routines such as insert, delete, and reverse.
AU - Alur, Rajeev
AU - Cerny, Pavol
ID - 3325
IS - 1
TI - Streaming transducers for algorithmic verification of single pass list processing programs
VL - 46
ER -
TY - CONF
AB - Weighted automata map input words to numerical values. Ap- plications of weighted automata include formal verification of quantitative properties, as well as text, speech, and image processing. A weighted au- tomaton is defined with respect to a semiring. For the tropical semiring, the weight of a run is the sum of the weights of the transitions taken along the run, and the value of a word is the minimal weight of an accepting run on it. In the 90’s, Krob studied the decidability of problems on rational series defined with respect to the tropical semiring. Rational series are strongly related to weighted automata, and Krob’s results apply to them. In par- ticular, it follows from Krob’s results that the universality problem (that is, deciding whether the values of all words are below some threshold) is decidable for weighted automata defined with respect to the tropical semir- ing with domain ∪ {∞}, and that the equality problem is undecidable when the domain is ∪ {∞}. In this paper we continue the study of the borders of decidability in weighted automata, describe alternative and direct proofs of the above results, and tighten them further. Unlike the proofs of Krob, which are algebraic in their nature, our proofs stay in the terrain of state machines, and the reduction is from the halting problem of a two-counter machine. This enables us to significantly simplify Krob’s reasoning, make the un- decidability result accessible to the automata-theoretic community, and strengthen it to apply already to a very simple class of automata: all the states are accepting, there are no initial nor final weights, and all the weights on the transitions are from the set {−1, 0, 1}. The fact we work directly with the automata enables us to tighten also the decidability re- sults and to show that the universality problem for weighted automata defined with respect to the tropical semiring with domain ∪ {∞}, and in fact even with domain ≥0 ∪ {∞}, is PSPACE-complete. Our results thus draw a sharper picture about the decidability of decision problems for weighted automata, in both the front of containment vs. universality and the front of the ∪ {∞} vs. the ∪ {∞} domains.
AU - Almagor, Shaull
AU - Boker, Udi
AU - Kupferman, Orna
ID - 3326
TI - What’s decidable about weighted automata
VL - 6996
ER -
TY - CONF
AB - We report on a generic uni- and bivariate algebraic kernel that is publicly available with CGAL 3.7. It comprises complete, correct, though efficient state-of-the-art implementations on polynomials, roots of polynomial systems, and the support to analyze algebraic curves defined by bivariate polynomials. The kernel design is generic, that is, various number types and substeps can be exchanged. It is accompanied with a ready-to-use interface to enable arrangements induced by algebraic curves, that have already been used as basis for various geometric applications, as arrangements on Dupin cyclides or the triangulation of algebraic surfaces. We present two novel applications: arrangements of rotated algebraic curves and Boolean set operations on polygons bounded by segments of algebraic curves. We also provide experiments showing that our general implementation is competitive and even often clearly outperforms existing implementations that are explicitly tailored for specific types of non-linear curves that are available in CGAL.
AU - Berberich, Eric
AU - Hemmer, Michael
AU - Kerber, Michael
ID - 3328
TI - A generic algebraic kernel for non linear geometric applications
ER -
TY - CONF
AB - Playing table tennis is a difficult task for robots, especially due to their limitations of acceleration. A key bottleneck is the amount of time needed to reach the desired hitting position and velocity of the racket for returning the incoming ball. Here, it often does not suffice to simply extrapolate the ball's trajectory after the opponent returns it but more information is needed. Humans are able to predict the ball's trajectory based on the opponent's moves and, thus, have a considerable advantage. Hence, we propose to incorporate an anticipation system into robot table tennis players, which enables the robot to react earlier while the opponent is performing the striking movement. Based on visual observation of the opponent's racket movement, the robot can predict the aim of the opponent and adjust its movement generation accordingly. The policies for deciding how and when to react are obtained by reinforcement learning. We conduct experiments with an existing robot player to show that the learned reaction policy can significantly improve the performance of the overall system.
AU - Wang, Zhikun
AU - Lampert, Christoph
AU - Mülling, Katharina
AU - Schölkopf, Bernhard
AU - Peters, Jan
ID - 3337
TI - Learning anticipation policies for robot table tennis
ER -
TY - CONF
AB - We study multi-label prediction for structured output sets, a problem that occurs, for example, in object detection in images, secondary structure prediction in computational biology, and graph matching with symmetries. Conventional multilabel classification techniques are typically not applicable in this situation, because they require explicit enumeration of the label set, which is infeasible in case of structured outputs. Relying on techniques originally designed for single-label structured prediction, in particular structured support vector machines, results in reduced prediction accuracy, or leads to infeasible optimization problems. In this work we derive a maximum-margin training formulation for multi-label structured prediction that remains computationally tractable while achieving high prediction accuracy. It also shares most beneficial properties with single-label maximum-margin approaches, in particular formulation as a convex optimization problem, efficient working set training, and PAC-Bayesian generalization bounds.
AU - Lampert, Christoph
ID - 3163
TI - Maximum margin multi-label structured prediction
ER -
TY - GEN
AB - We study multi-label prediction for structured output spaces, a problem that occurs, for example, in object detection in images, secondary structure prediction in computational biology, and graph matching with symmetries. Conventional multi-label classification techniques are typically not applicable in this situation, because they require explicit enumeration of the label space, which is infeasible in case of structured outputs. Relying on techniques originally designed for single- label structured prediction, in particular structured support vector machines, results in reduced prediction accuracy, or leads to infeasible optimization problems. In this work we derive a maximum-margin training formulation for multi-label structured prediction that remains computationally tractable while achieving high prediction accuracy. It also shares most beneficial properties with single-label maximum-margin approaches, in particular a formulation as a convex optimization problem, efficient working set training, and PAC-Bayesian generalization bounds.
AU - Lampert, Christoph
ID - 3322
T2 - NIPS: Neural Information Processing Systems
TI - Maximum margin multi label structured prediction
ER -
TY - CONF
AB - We study Markov decision processes (MDPs) with multiple limit-average (or mean-payoff) functions. We consider two different objectives, namely, expectation and satisfaction objectives. Given an MDP with k reward functions, in the expectation objective the goal is to maximize the expected limit-average value, and in the satisfaction objective the goal is to maximize the probability of runs such that the limit-average value stays above a given vector. We show that under the expectation objective, in contrast to the single-objective case, both randomization and memory are necessary for strategies, and that finite-memory randomized strategies are sufficient. Under the satisfaction objective, in contrast to the single-objective case, infinite memory is necessary for strategies, and that randomized memoryless strategies are sufficient for epsilon-approximation, for all epsilon>;0. We further prove that the decision problems for both expectation and satisfaction objectives can be solved in polynomial time and the trade-off curve (Pareto curve) can be epsilon-approximated in time polynomial in the size of the MDP and 1/epsilon, and exponential in the number of reward functions, for all epsilon>;0. Our results also reveal flaws in previous work for MDPs with multiple mean-payoff functions under the expectation objective, correct the flaws and obtain improved results.
AU - Brázdil, Tomáš
AU - Brožek, Václav
AU - Chatterjee, Krishnendu
AU - Forejt, Vojtěch
AU - Kučera, Antonín
ID - 3346
TI - Two views on multiple mean payoff objectives in Markov Decision Processes
ER -
TY - CONF
AB - The class of omega-regular languages provides a robust specification language in verification. Every omega-regular condition can be decomposed into a safety part and a liveness part. The liveness part ensures that something good happens "eventually". Finitary liveness was proposed by Alur and Henzinger as a stronger formulation of liveness. It requires that there exists an unknown, fixed bound b such that something good happens within b transitions. In this work we consider automata with finitary acceptance conditions defined by finitary Buchi, parity and Streett languages. We study languages expressible by such automata: we give their topological complexity and present a regular-expression characterization. We compare the expressive power of finitary automata and give optimal algorithms for classical decisions questions. We show that the finitary languages are Sigma 2-complete; we present a complete picture of the expressive power of various classes of automata with finitary and infinitary acceptance conditions; we show that the languages defined by finitary parity automata exactly characterize the star-free fragment of omega B-regular languages; and we show that emptiness is NLOGSPACE-complete and universality as well as language inclusion are PSPACE-complete for finitary parity and Streett automata.
AU - Chatterjee, Krishnendu
AU - Fijalkow, Nathanaël
ID - 3347
TI - Finitary languages
VL - 6638
ER -
TY - CONF
AB - We study synthesis of controllers for real-time systems, where the objective is to stay in a given safe set. The problem is solved by obtaining winning strategies in the setting of concurrent two-player timed automaton games with safety objectives. To prevent a player from winning by blocking time, we restrict each player to strategies that ensure that the player cannot be responsible for causing a zeno run. We construct winning strategies for the controller which require access only to (1) the system clocks (thus, controllers which require their own internal infinitely precise clocks are not necessary), and (2) a linear (in the number of clocks) number of memory bits. Precisely, we show that for safety objectives, a memory of size (3 · |C|+lg(|C|+1)) bits suffices for winning controller strategies, where C is the set of clocks of the timed automaton game, significantly improving the previous known exponential bound. We also settle the open question of whether winning region controller strategies require memory for safety objectives by showing with an example the necessity of memory for region strategies to win for safety objectives.
AU - Chatterjee, Krishnendu
AU - Prabhu, Vinayak
ID - 3348
TI - Synthesis of memory efficient real time controllers for safety objectives
ER -
TY - CONF
AB - Games on graphs provide a natural model for reactive non-terminating systems. In such games, the interaction of two players on an arena results in an infinite path that describes a run of the system. Different settings are used to model various open systems in computer science, as for instance turn-based or concurrent moves, and deterministic or stochastic transitions. In this paper, we are interested in turn-based games, and specifically in deterministic parity games and stochastic reachability games (also known as simple stochastic games). We present a simple, direct and efficient reduction from deterministic parity games to simple stochastic games: it yields an arena whose size is linear up to a logarithmic factor in size of the original arena.
AU - Chatterjee, Krishnendu
AU - Fijalkow, Nathanaël
ID - 3349
TI - A reduction from parity games to simple stochastic games
VL - 54
ER -
TY - CONF
AB - A controller for a discrete game with ω-regular objectives requires attention if, intuitively, it requires measuring the state and switching from the current control action. Minimum attention controllers are preferable in modern shared implementations of cyber-physical systems because they produce the least burden on system resources such as processor time or communication bandwidth. We give algorithms to compute minimum attention controllers for ω-regular objectives in imperfect information discrete two-player games. We show a polynomial-time reduction from minimum attention controller synthesis to synthesis of controllers for mean-payoff parity objectives in games of incomplete information. This gives an optimal EXPTIME-complete synthesis algorithm. We show that the minimum attention controller problem is decidable for infinite state systems with finite bisimulation quotients. In particular, the problem is decidable for timed and rectangular automata.
AU - Chatterjee, Krishnendu
AU - Majumdar, Ritankar
ED - Fahrenberg, Uli
ED - Tripakis, Stavros
ID - 3350
TI - Minimum attention controller synthesis for omega regular objectives
VL - 6919
ER -
TY - CONF
AB - In two-player games on graph, the players construct an infinite path through the game graph and get a reward computed by a payoff function over infinite paths. Over weighted graphs, the typical and most studied payoff functions compute the limit-average or the discounted sum of the rewards along the path. Besides their simple definition, these two payoff functions enjoy the property that memoryless optimal strategies always exist. In an attempt to construct other simple payoff functions, we define a class of payoff functions which compute an (infinite) weighted average of the rewards. This new class contains both the limit-average and the discounted sum functions, and we show that they are the only members of this class which induce memoryless optimal strategies, showing that there is essentially no other simple payoff functions.
AU - Chatterjee, Krishnendu
AU - Doyen, Laurent
AU - Singh, Rohit
ED - Owe, Olaf
ED - Steffen, Martin
ED - Telle, Jan Arne
ID - 3351
TI - On memoryless quantitative objectives
VL - 6914
ER -
TY - JOUR
AB - Exploring the connection of biology with reactive systems to better understand living systems.
AU - Fisher, Jasmin
AU - Harel, David
AU - Henzinger, Thomas A
ID - 3352
IS - 10
JF - Communications of the ACM
TI - Biology as reactivity
VL - 54
ER -
TY - JOUR
AB - Compositional theories are crucial when designing large and complex systems from smaller components. In this work we propose such a theory for synchronous concurrent systems. Our approach follows so-called interface theories, which use game-theoretic interpretations of composition and refinement. These are appropriate for systems with distinct inputs and outputs, and explicit conditions on inputs that must be enforced during composition. Our interfaces model systems that execute in an infinite sequence of synchronous rounds. At each round, a contract must be satisfied. The contract is simply a relation specifying the set of valid input/output pairs. Interfaces can be composed by parallel, serial or feedback composition. A refinement relation between interfaces is defined, and shown to have two main properties: (1) it is preserved by composition, and (2) it is equivalent to substitutability, namely, the ability to replace an interface by another one in any context. Shared refinement and abstraction operators, corresponding to greatest lower and least upper bounds with respect to refinement, are also defined. Input-complete interfaces, that impose no restrictions on inputs, and deterministic interfaces, that produce a unique output for any legal input, are discussed as special cases, and an interesting duality between the two classes is exposed. A number of illustrative examples are provided, as well as algorithms to compute compositions, check refinement, and so on, for finite-state interfaces.
AU - Tripakis, Stavros
AU - Lickly, Ben
AU - Henzinger, Thomas A
AU - Lee, Edward
ID - 3353
IS - 4
JF - ACM Transactions on Programming Languages and Systems (TOPLAS)
TI - A theory of synchronous relational interfaces
VL - 33
ER -
TY - GEN
AB - Turn-based stochastic games and its important subclass Markov decision processes (MDPs) provide models for systems with both probabilistic and nondeterministic behaviors. We consider turn-based stochastic games with two classical quantitative objectives: discounted-sum and long-run average objectives. The game models and the quantitative objectives are widely used in probabilistic verification, planning, optimal inventory control, network protocol and performance analysis. Games and MDPs that model realistic systems often have very large state spaces, and probabilistic abstraction techniques are necessary to handle the state-space explosion. The commonly used full-abstraction techniques do not yield space-savings for systems that have many states with similar value, but does not necessarily have similar transition structure. A semi-abstraction technique, namely Magnifying-lens abstractions (MLA), that clusters states based on value only, disregarding differences in their transition relation was proposed for qualitative objectives (reachability and safety objectives). In this paper we extend the MLA technique to solve stochastic games with discounted-sum and long-run average objectives. We present the MLA technique based abstraction-refinement algorithm for stochastic games and MDPs with discounted-sum objectives. For long-run average objectives, our solution works for all MDPs and a sub-class of stochastic games where every state has the same value.
AU - Chatterjee, Krishnendu
AU - De Alfaro, Luca
AU - Pritam, Roy
ID - 3339
T2 - arXiv
TI - Magnifying lens abstraction for stochastic games with discounted and long-run average objectives
ER -
TY - CONF
AB - We consider Markov decision processes (MDPs) with ω-regular specifications given as parity objectives. We consider the problem of computing the set of almost-sure winning states from where the objective can be ensured with probability 1. The algorithms for the computation of the almost-sure winning set for parity objectives iteratively use the solutions for the almost-sure winning set for Büchi objectives (a special case of parity objectives). Our contributions are as follows: First, we present the first subquadratic symbolic algorithm to compute the almost-sure winning set for MDPs with Büchi objectives; our algorithm takes O(nm) symbolic steps as compared to the previous known algorithm that takes O(n 2) symbolic steps, where n is the number of states and m is the number of edges of the MDP. In practice MDPs often have constant out-degree, and then our symbolic algorithm takes O(nn) symbolic steps, as compared to the previous known O(n 2) symbolic steps algorithm. Second, we present a new algorithm, namely win-lose algorithm, with the following two properties: (a) the algorithm iteratively computes subsets of the almost-sure winning set and its complement, as compared to all previous algorithms that discover the almost-sure winning set upon termination; and (b) requires O(nK) symbolic steps, where K is the maximal number of edges of strongly connected components (scc’s) of the MDP. The win-lose algorithm requires symbolic computation of scc’s. Third, we improve the algorithm for symbolic scc computation; the previous known algorithm takes linear symbolic steps, and our new algorithm improves the constants associated with the linear number of steps. In the worst case the previous known algorithm takes 5·n symbolic steps, whereas our new algorithm takes 4 ·n symbolic steps.
AU - Chatterjee, Krishnendu
AU - Henzinger, Monika
AU - Joglekar, Manas
AU - Nisarg, Shah
ED - Gopalakrishnan, Ganesh
ED - Qadeer, Shaz
ID - 3342
TI - Symbolic algorithms for qualitative analysis of Markov decision processes with Büchi objectives
VL - 6806
ER -
TY - CONF
AB - We present faster and dynamic algorithms for the following problems arising in probabilistic verification: Computation of the maximal end-component (mec) decomposition of Markov decision processes (MDPs), and of the almost sure winning set for reachability and parity objectives in MDPs. We achieve the following running time for static algorithms in MDPs with graphs of n vertices and m edges: (1) O(m · min{ √m, n2/3 }) for the mec decomposition, improving the longstanding O(m·n) bound; (2) O(m·n2/3) for reachability objectives, improving the previous O(m · √m) bound for m > n4/3; and (3) O(m · min{ √m, n2/3 } · log(d)) for parity objectives with d priorities, improving the previous O(m · √m · d) bound. We also give incremental and decremental algorithms in linear time for mec decomposition and reachability objectives and O(m · log d) time for parity ob jectives.
AU - Chatterjee, Krishnendu
AU - Henzinger, Monika
ID - 3343
TI - Faster and dynamic algorithms for maximal end component decomposition and related graph problems in probabilistic verification
ER -
TY - CONF
AB - Games played on graphs provide the mathematical framework to analyze several important problems in computer science as well as mathematics, such as the synthesis problem of Church, model checking of open reactive systems and many others. On the basis of mode of interaction of the players these games can be classified as follows: (a) turn-based (players make moves in turns); and (b) concurrent (players make moves simultaneously). On the basis of the information available to the players these games can be classified as follows: (a) perfect-information (players have perfect view of the game); and (b) partial-information (players have partial view of the game). In this talk we will consider all these classes of games with reachability objectives, where the goal of one player is to reach a set of target vertices of the graph, and the goal of the opponent player is to prevent the player from reaching the target. We will survey the results for various classes of games, and the results range from linear time decision algorithms to EXPTIME-complete problems to undecidable problems.
AU - Chatterjee, Krishnendu
ED - Delzanno, Giorgo
ED - Potapov, Igor
ID - 3344
TI - Graph games with reachability objectives
VL - 6945
ER -