@inproceedings{10665, abstract = {Formal verification of neural networks is an active topic of research, and recent advances have significantly increased the size of the networks that verification tools can handle. However, most methods are designed for verification of an idealized model of the actual network which works over real arithmetic and ignores rounding imprecisions. This idealization is in stark contrast to network quantization, which is a technique that trades numerical precision for computational efficiency and is, therefore, often applied in practice. Neglecting rounding errors of such low-bit quantized neural networks has been shown to lead to wrong conclusions about the network’s correctness. Thus, the desired approach for verifying quantized neural networks would be one that takes these rounding errors into account. In this paper, we show that verifying the bitexact implementation of quantized neural networks with bitvector specifications is PSPACE-hard, even though verifying idealized real-valued networks and satisfiability of bit-vector specifications alone are each in NP. Furthermore, we explore several practical heuristics toward closing the complexity gap between idealized and bit-exact verification. In particular, we propose three techniques for making SMT-based verification of quantized neural networks more scalable. Our experiments demonstrate that our proposed methods allow a speedup of up to three orders of magnitude over existing approaches.}, author = {Henzinger, Thomas A and Lechner, Mathias and Zikelic, Dorde}, booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence}, isbn = {978-1-57735-866-4}, issn = {2374-3468}, location = {Virtual}, number = {5A}, pages = {3787--3795}, publisher = {AAAI Press}, title = {{Scalable verification of quantized neural networks}}, volume = {35}, year = {2021}, } @inproceedings{10668, abstract = {Robustness to variations in lighting conditions is a key objective for any deep vision system. To this end, our paper extends the receptive field of convolutional neural networks with two residual components, ubiquitous in the visual processing system of vertebrates: On-center and off-center pathways, with an excitatory center and inhibitory surround; OOCS for short. The On-center pathway is excited by the presence of a light stimulus in its center, but not in its surround, whereas the Off-center pathway is excited by the absence of a light stimulus in its center, but not in its surround. We design OOCS pathways via a difference of Gaussians, with their variance computed analytically from the size of the receptive fields. OOCS pathways complement each other in their response to light stimuli, ensuring this way a strong edge-detection capability, and as a result an accurate and robust inference under challenging lighting conditions. We provide extensive empirical evidence showing that networks supplied with OOCS pathways gain accuracy and illumination-robustness from the novel edge representation, compared to other baselines.}, author = {Babaiee, Zahra and Hasani, Ramin and Lechner, Mathias and Rus, Daniela and Grosu, Radu}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, issn = {2640-3498}, location = {Virtual}, pages = {478--489}, publisher = {ML Research Press}, title = {{On-off center-surround receptive fields for accurate and robust image classification}}, volume = {139}, year = {2021}, } @inproceedings{10669, abstract = {We show that Neural ODEs, an emerging class of timecontinuous neural networks, can be verified by solving a set of global-optimization problems. For this purpose, we introduce Stochastic Lagrangian Reachability (SLR), an abstraction-based technique for constructing a tight Reachtube (an over-approximation of the set of reachable states over a given time-horizon), and provide stochastic guarantees in the form of confidence intervals for the Reachtube bounds. SLR inherently avoids the infamous wrapping effect (accumulation of over-approximation errors) by performing local optimization steps to expand safe regions instead of repeatedly forward-propagating them as is done by deterministic reachability methods. To enable fast local optimizations, we introduce a novel forward-mode adjoint sensitivity method to compute gradients without the need for backpropagation. Finally, we establish asymptotic and non-asymptotic convergence rates for SLR.}, author = {Grunbacher, Sophie and Hasani, Ramin and Lechner, Mathias and Cyranka, Jacek and Smolka, Scott A and Grosu, Radu}, booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence}, isbn = {978-1-57735-866-4}, issn = {2374-3468}, location = {Virtual}, number = {13}, pages = {11525--11535}, publisher = {AAAI Press}, title = {{On the verification of neural ODEs with stochastic guarantees}}, volume = {35}, year = {2021}, } @inproceedings{10671, abstract = {We introduce a new class of time-continuous recurrent neural network models. Instead of declaring a learning system’s dynamics by implicit nonlinearities, we construct networks of linear first-order dynamical systems modulated via nonlinear interlinked gates. The resulting models represent dynamical systems with varying (i.e., liquid) time-constants coupled to their hidden state, with outputs being computed by numerical differential equation solvers. These neural networks exhibit stable and bounded behavior, yield superior expressivity within the family of neural ordinary differential equations, and give rise to improved performance on time-series prediction tasks. To demonstrate these properties, we first take a theoretical approach to find bounds over their dynamics, and compute their expressive power by the trajectory length measure in a latent trajectory space. We then conduct a series of time-series prediction experiments to manifest the approximation capability of Liquid Time-Constant Networks (LTCs) compared to classical and modern RNNs.}, author = {Hasani, Ramin and Lechner, Mathias and Amini, Alexander and Rus, Daniela and Grosu, Radu}, booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence}, isbn = {978-1-57735-866-4}, issn = {2374-3468}, location = {Virtual}, number = {9}, pages = {7657--7666}, publisher = {AAAI Press}, title = {{Liquid time-constant networks}}, volume = {35}, year = {2021}, } @inproceedings{10688, abstract = {Civl is a static verifier for concurrent programs designed around the conceptual framework of layered refinement, which views the task of verifying a program as a sequence of program simplification steps each justified by its own invariant. Civl verifies a layered concurrent program that compactly expresses all the programs in this sequence and the supporting invariants. This paper presents the design and implementation of the Civl verifier.}, author = {Kragl, Bernhard and Qadeer, Shaz}, booktitle = {Proceedings of the 21st Conference on Formal Methods in Computer-Aided Design}, editor = {Ruzica, Piskac and Whalen, Michael W.}, isbn = {978-3-85448-046-4}, location = {Virtual}, pages = {143–152}, publisher = {TU Wien Academic Press}, title = {{The Civl verifier}}, doi = {10.34727/2021/isbn.978-3-85448-046-4_23}, volume = {2}, year = {2021}, } @article{8912, abstract = {For automata, synchronization, the problem of bringing an automaton to a particular state regardless of its initial state, is important. It has several applications in practice and is related to a fifty-year-old conjecture on the length of the shortest synchronizing word. Although using shorter words increases the effectiveness in practice, finding a shortest one (which is not necessarily unique) is NP-hard. For this reason, there exist various heuristics in the literature. However, high-quality heuristics such as SynchroP producing relatively shorter sequences are very expensive and can take hours when the automaton has tens of thousands of states. The SynchroP heuristic has been frequently used as a benchmark to evaluate the performance of the new heuristics. In this work, we first improve the runtime of SynchroP and its variants by using algorithmic techniques. We then focus on adapting SynchroP for many-core architectures, and overall, we obtain more than 1000× speedup on GPUs compared to naive sequential implementation that has been frequently used as a benchmark to evaluate new heuristics in the literature. We also propose two SynchroP variants and evaluate their performance.}, author = {Sarac, Naci E and Altun, Ömer Faruk and Atam, Kamil Tolga and Karahoda, Sertac and Kaya, Kamer and Yenigün, Hüsnü}, issn = {1873-6793}, journal = {Expert Systems with Applications}, number = {4}, publisher = {Elsevier}, title = {{Boosting expensive synchronizing heuristics}}, doi = {10.1016/j.eswa.2020.114203}, volume = {167}, year = {2021}, } @inproceedings{9200, abstract = {Formal design of embedded and cyber-physical systems relies on mathematical modeling. In this paper, we consider the model class of hybrid automata whose dynamics are defined by affine differential equations. Given a set of time-series data, we present an algorithmic approach to synthesize a hybrid automaton exhibiting behavior that is close to the data, up to a specified precision, and changes in synchrony with the data. A fundamental problem in our synthesis algorithm is to check membership of a time series in a hybrid automaton. Our solution integrates reachability and optimization techniques for affine dynamical systems to obtain both a sufficient and a necessary condition for membership, combined in a refinement framework. The algorithm processes one time series at a time and hence can be interrupted, provide an intermediate result, and be resumed. We report experimental results demonstrating the applicability of our synthesis approach.}, author = {Garcia Soto, Miriam and Henzinger, Thomas A and Schilling, Christian}, booktitle = {HSCC '21: Proceedings of the 24th International Conference on Hybrid Systems: Computation and Control}, isbn = {9781450383394}, keywords = {hybrid automaton, membership, system identification}, location = {Nashville, TN, United States}, pages = {2102.12734}, publisher = {Association for Computing Machinery}, title = {{Synthesis of hybrid automata with affine dynamics from time-series data}}, doi = {10.1145/3447928.3456704}, year = {2021}, } @article{9239, abstract = {A graph game proceeds as follows: two players move a token through a graph to produce a finite or infinite path, which determines the payoff of the game. We study bidding games in which in each turn, an auction determines which player moves the token. Bidding games were largely studied in combination with two variants of first-price auctions called “Richman” and “poorman” bidding. We study taxman bidding, which span the spectrum between the two. The game is parameterized by a constant : portion τ of the winning bid is paid to the other player, and portion to the bank. While finite-duration (reachability) taxman games have been studied before, we present, for the first time, results on infinite-duration taxman games: we unify, generalize, and simplify previous equivalences between bidding games and a class of stochastic games called random-turn games.}, author = {Avni, Guy and Henzinger, Thomas A and Žikelić, Đorđe}, issn = {1090-2724}, journal = {Journal of Computer and System Sciences}, number = {8}, pages = {133--144}, publisher = {Elsevier}, title = {{Bidding mechanisms in graph games}}, doi = {10.1016/j.jcss.2021.02.008}, volume = {119}, year = {2021}, } @unpublished{9281, abstract = {We comment on two formal proofs of Fermat's sum of two squares theorem, written using the Mathematical Components libraries of the Coq proof assistant. The first one follows Zagier's celebrated one-sentence proof; the second follows David Christopher's recent new proof relying on partition-theoretic arguments. Both formal proofs rely on a general property of involutions of finite sets, of independent interest. The proof technique consists for the most part of automating recurrent tasks (such as case distinctions and computations on natural numbers) via ad hoc tactics.}, author = {Dubach, Guillaume and Mühlböck, Fabian}, booktitle = {arXiv}, title = {{Formal verification of Zagier's one-sentence proof}}, doi = {10.48550/arXiv.2103.11389}, year = {2021}, } @article{9647, abstract = {Gene expression is regulated by the set of transcription factors (TFs) that bind to the promoter. The ensuing regulating function is often represented as a combinational logic circuit, where output (gene expression) is determined by current input values (promoter bound TFs) only. However, the simultaneous arrival of TFs is a strong assumption, since transcription and translation of genes introduce intrinsic time delays and there is no global synchronisation among the arrival times of different molecular species at their targets. We present an experimentally implementable genetic circuit with two inputs and one output, which in the presence of small delays in input arrival, exhibits qualitatively distinct population-level phenotypes, over timescales that are longer than typical cell doubling times. From a dynamical systems point of view, these phenotypes represent long-lived transients: although they converge to the same value eventually, they do so after a very long time span. The key feature of this toy model genetic circuit is that, despite having only two inputs and one output, it is regulated by twenty-three distinct DNA-TF configurations, two of which are more stable than others (DNA looped states), one promoting and another blocking the expression of the output gene. Small delays in input arrival time result in a majority of cells in the population quickly reaching the stable state associated with the first input, while exiting of this stable state occurs at a slow timescale. In order to mechanistically model the behaviour of this genetic circuit, we used a rule-based modelling language, and implemented a grid-search to find parameter combinations giving rise to long-lived transients. Our analysis shows that in the absence of feedback, there exist path-dependent gene regulatory mechanisms based on the long timescale of transients. The behaviour of this toy model circuit suggests that gene regulatory networks can exploit event timing to create phenotypes, and it opens the possibility that they could use event timing to memorise events, without regulatory feedback. The model reveals the importance of (i) mechanistically modelling the transitions between the different DNA-TF states, and (ii) employing transient analysis thereof.}, author = {Petrov, Tatjana and Igler, Claudia and Sezgin, Ali and Henzinger, Thomas A and Guet, Calin C}, issn = {0304-3975}, journal = {Theoretical Computer Science}, pages = {1--16}, publisher = {Elsevier}, title = {{Long lived transients in gene regulation}}, doi = {10.1016/j.tcs.2021.05.023}, volume = {893}, year = {2021}, } @misc{9946, abstract = {We argue that the time is ripe to investigate differential monitoring, in which the specification of a program's behavior is implicitly given by a second program implementing the same informal specification. Similar ideas have been proposed before, and are currently implemented in restricted form for testing and specialized run-time analyses, aspects of which we combine. We discuss the challenges of implementing differential monitoring as a general-purpose, black-box run-time monitoring framework, and present promising results of a preliminary implementation, showing low monitoring overheads for diverse programs.}, author = {Mühlböck, Fabian and Henzinger, Thomas A}, issn = {2664-1690}, keywords = {run-time verification, software engineering, implicit specification}, pages = {17}, publisher = {IST Austria}, title = {{Differential monitoring}}, doi = {10.15479/AT:ISTA:9946}, year = {2021}, } @article{10674, abstract = {In two-player games on graphs, the players move a token through a graph to produce an infinite path, which determines the winner of the game. Such games are central in formal methods since they model the interaction between a non-terminating system and its environment. In bidding games the players bid for the right to move the token: in each round, the players simultaneously submit bids, and the higher bidder moves the token and pays the other player. Bidding games are known to have a clean and elegant mathematical structure that relies on the ability of the players to submit arbitrarily small bids. Many applications, however, require a fixed granularity for the bids, which can represent, for example, the monetary value expressed in cents. We study, for the first time, the combination of discrete-bidding and infinite-duration games. Our most important result proves that these games form a large determined subclass of concurrent games, where determinacy is the strong property that there always exists exactly one player who can guarantee winning the game. In particular, we show that, in contrast to non-discrete bidding games, the mechanism with which tied bids are resolved plays an important role in discrete-bidding games. We study several natural tie-breaking mechanisms and show that, while some do not admit determinacy, most natural mechanisms imply determinacy for every pair of initial budgets.}, author = {Aghajohari, Milad and Avni, Guy and Henzinger, Thomas A}, issn = {1860-5974}, journal = {Logical Methods in Computer Science}, keywords = {computer science, computer science and game theory, logic in computer science}, number = {1}, pages = {10:1--10:23}, publisher = {EPI Sciences}, title = {{Determinacy in discrete-bidding infinite-duration games}}, doi = {10.23638/LMCS-17(1:10)2021}, volume = {17}, year = {2021}, } @inproceedings{10666, abstract = {Adversarial training is an effective method to train deep learning models that are resilient to norm-bounded perturbations, with the cost of nominal performance drop. While adversarial training appears to enhance the robustness and safety of a deep model deployed in open-world decision-critical applications, counterintuitively, it induces undesired behaviors in robot learning settings. In this paper, we show theoretically and experimentally that neural controllers obtained via adversarial training are subjected to three types of defects, namely transient, systematic, and conditional errors. We first generalize adversarial training to a safety-domain optimization scheme allowing for more generic specifications. We then prove that such a learning process tends to cause certain error profiles. We support our theoretical results by a thorough experimental safety analysis in a robot-learning task. Our results suggest that adversarial training is not yet ready for robot learning.}, author = {Lechner, Mathias and Hasani, Ramin and Grosu, Radu and Rus, Daniela and Henzinger, Thomas A}, booktitle = {2021 IEEE International Conference on Robotics and Automation}, isbn = {978-1-7281-9078-5}, issn = {2577-087X}, location = {Xi'an, China}, pages = {4140--4147}, publisher = {IEEE}, title = {{Adversarial training is not ready for robot learning}}, doi = {10.1109/ICRA48506.2021.9561036}, year = {2021}, } @inproceedings{10667, abstract = {Bayesian neural networks (BNNs) place distributions over the weights of a neural network to model uncertainty in the data and the network's prediction. We consider the problem of verifying safety when running a Bayesian neural network policy in a feedback loop with infinite time horizon systems. Compared to the existing sampling-based approaches, which are inapplicable to the infinite time horizon setting, we train a separate deterministic neural network that serves as an infinite time horizon safety certificate. In particular, we show that the certificate network guarantees the safety of the system over a subset of the BNN weight posterior's support. Our method first computes a safe weight set and then alters the BNN's weight posterior to reject samples outside this set. Moreover, we show how to extend our approach to a safe-exploration reinforcement learning setting, in order to avoid unsafe trajectories during the training of the policy. We evaluate our approach on a series of reinforcement learning benchmarks, including non-Lyapunovian safety specifications.}, author = {Lechner, Mathias and Žikelić, Ðorđe and Chatterjee, Krishnendu and Henzinger, Thomas A}, booktitle = {35th Conference on Neural Information Processing Systems}, issn = {1049-5258}, location = {Virtual}, publisher = {Neural Information Processing Systems Foundation}, title = {{Infinite time horizon safety of Bayesian neural networks}}, doi = {10.48550/arXiv.2111.03165}, year = {2021}, } @inproceedings{10670, abstract = {Imitation learning enables high-fidelity, vision-based learning of policies within rich, photorealistic environments. However, such techniques often rely on traditional discrete-time neural models and face difficulties in generalizing to domain shifts by failing to account for the causal relationships between the agent and the environment. In this paper, we propose a theoretical and experimental framework for learning causal representations using continuous-time neural networks, specifically over their discrete-time counterparts. We evaluate our method in the context of visual-control learning of drones over a series of complex tasks, ranging from short- and long-term navigation, to chasing static and dynamic objects through photorealistic environments. Our results demonstrate that causal continuous-time deep models can perform robust navigation tasks, where advanced recurrent models fail. These models learn complex causal control representations directly from raw visual inputs and scale to solve a variety of tasks using imitation learning.}, author = {Vorbach, Charles J and Hasani, Ramin and Amini, Alexander and Lechner, Mathias and Rus, Daniela}, booktitle = {35th Conference on Neural Information Processing Systems}, issn = {1049-5258}, location = {Virtual}, publisher = {Neural Information Processing Systems Foundation}, title = {{Causal navigation by continuous-time neural networks}}, year = {2021}, } @inproceedings{9356, abstract = {In runtime verification, a monitor watches a trace of a system and, if possible, decides after observing each finite prefix whether or not the unknown infinite trace satisfies a given specification. We generalize the theory of runtime verification to monitors that attempt to estimate numerical values of quantitative trace properties (instead of attempting to conclude boolean values of trace specifications), such as maximal or average response time along a trace. Quantitative monitors are approximate: with every finite prefix, they can improve their estimate of the infinite trace's unknown property value. Consequently, quantitative monitors can be compared with regard to a precision-cost trade-off: better approximations of the property value require more monitor resources, such as states (in the case of finite-state monitors) or registers, and additional resources yield better approximations. We introduce a formal framework for quantitative and approximate monitoring, show how it conservatively generalizes the classical boolean setting for monitoring, and give several precision-cost trade-offs for monitors. For example, we prove that there are quantitative properties for which every additional register improves monitoring precision.}, author = {Henzinger, Thomas A and Sarac, Naci E}, booktitle = {Proceedings of the 36th Annual ACM/IEEE Symposium on Logic in Computer Science}, location = {Online}, publisher = {Institute of Electrical and Electronics Engineers}, title = {{Quantitative and approximate monitoring}}, doi = {10.1109/LICS52264.2021.9470547}, year = {2021}, } @article{10861, abstract = {We introduce in this paper AMT2.0, a tool for qualitative and quantitative analysis of hybrid continuous and Boolean signals that combine numerical values and discrete events. The evaluation of the signals is based on rich temporal specifications expressed in extended signal temporal logic, which integrates timed regular expressions within signal temporal logic. The tool features qualitative monitoring (property satisfaction checking), trace diagnostics for explaining and justifying property violations and specification-driven measurement of quantitative features of the signal. We demonstrate the tool functionality on several running examples and case studies, and evaluate its performance.}, author = {Nickovic, Dejan and Lebeltel, Olivier and Maler, Oded and Ferrere, Thomas and Ulus, Dogan}, issn = {1433-2787}, journal = {International Journal on Software Tools for Technology Transfer}, keywords = {Information Systems, Software}, number = {6}, pages = {741--758}, publisher = {Springer Nature}, title = {{AMT 2.0: Qualitative and quantitative trace analysis with extended signal temporal logic}}, doi = {10.1007/s10009-020-00582-z}, volume = {22}, year = {2020}, } @inproceedings{10672, abstract = {The family of feedback alignment (FA) algorithms aims to provide a more biologically motivated alternative to backpropagation (BP), by substituting the computations that are unrealistic to be implemented in physical brains. While FA algorithms have been shown to work well in practice, there is a lack of rigorous theory proofing their learning capabilities. Here we introduce the first feedback alignment algorithm with provable learning guarantees. In contrast to existing work, we do not require any assumption about the size or depth of the network except that it has a single output neuron, i.e., such as for binary classification tasks. We show that our FA algorithm can deliver its theoretical promises in practice, surpassing the learning performance of existing FA methods and matching backpropagation in binary classification tasks. Finally, we demonstrate the limits of our FA variant when the number of output neurons grows beyond a certain quantity.}, author = {Lechner, Mathias}, booktitle = {8th International Conference on Learning Representations}, location = {Virtual ; Addis Ababa, Ethiopia}, publisher = {ICLR}, title = {{Learning representations for binary-classification without backpropagation}}, year = {2020}, } @inproceedings{10673, abstract = {We propose a neural information processing system obtained by re-purposing the function of a biological neural circuit model to govern simulated and real-world control tasks. Inspired by the structure of the nervous system of the soil-worm, C. elegans, we introduce ordinary neural circuits (ONCs), defined as the model of biological neural circuits reparameterized for the control of alternative tasks. We first demonstrate that ONCs realize networks with higher maximum flow compared to arbitrary wired networks. We then learn instances of ONCs to control a series of robotic tasks, including the autonomous parking of a real-world rover robot. For reconfiguration of the purpose of the neural circuit, we adopt a search-based optimization algorithm. Ordinary neural circuits perform on par and, in some cases, significantly surpass the performance of contemporary deep learning models. ONC networks are compact, 77% sparser than their counterpart neural controllers, and their neural dynamics are fully interpretable at the cell-level.}, author = {Hasani, Ramin and Lechner, Mathias and Amini, Alexander and Rus, Daniela and Grosu, Radu}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, issn = {2640-3498}, location = {Virtual}, pages = {4082--4093}, title = {{A natural lottery ticket winner: Reinforcement learning with ordinary neural circuits}}, year = {2020}, } @inproceedings{7808, abstract = {Quantization converts neural networks into low-bit fixed-point computations which can be carried out by efficient integer-only hardware, and is standard practice for the deployment of neural networks on real-time embedded devices. However, like their real-numbered counterpart, quantized networks are not immune to malicious misclassification caused by adversarial attacks. We investigate how quantization affects a network’s robustness to adversarial attacks, which is a formal verification question. We show that neither robustness nor non-robustness are monotonic with changing the number of bits for the representation and, also, neither are preserved by quantization from a real-numbered network. For this reason, we introduce a verification method for quantized neural networks which, using SMT solving over bit-vectors, accounts for their exact, bit-precise semantics. We built a tool and analyzed the effect of quantization on a classifier for the MNIST dataset. We demonstrate that, compared to our method, existing methods for the analysis of real-numbered networks often derive false conclusions about their quantizations, both when determining robustness and when detecting attacks, and that existing methods for quantized networks often miss attacks. Furthermore, we applied our method beyond robustness, showing how the number of bits in quantization enlarges the gender bias of a predictor for students’ grades.}, author = {Giacobbe, Mirco and Henzinger, Thomas A and Lechner, Mathias}, booktitle = {International Conference on Tools and Algorithms for the Construction and Analysis of Systems}, isbn = {9783030452360}, issn = {1611-3349}, location = {Dublin, Ireland}, pages = {79--97}, publisher = {Springer Nature}, title = {{How many bits does it take to quantize your neural network?}}, doi = {10.1007/978-3-030-45237-7_5}, volume = {12079}, year = {2020}, }