Stefan Mitsch

Samuel Teuber, Stefan Mitsch, André Platzer · cmu

While neural networks (NNs) have potential as autonomous controllers for Cyber-Physical Systems, verifying the safety of NN based control systems (NNCSs) poses significant challenges for the practical use of NNs, especially when safety is needed for unbounded time horizons. One reason is the intractability of analyzing NNs, ODEs and hybrid systems. To this end, we introduce VerSAILLE (Verifiably Safe AI via Logically Linked Envelopes): The first general approach that allows reusing control theory results for NNCS verification. By joining forces, we exploit the efficiency of NN verification tools while retaining the rigor of differential dynamic logic (dL). Based on provably safe control envelopes in dL, we derive specifications for the NN which is proven via NN verification. We show that a proof of the NN adhering to the specification is mirrored by a dL proof on the infinite-time safety of the NNCS. The NN verification properties resulting from hybrid systems typically contain nonlinear arithmetic and arbitrary logical structures while efficient NN verification merely supports linear constraints. To overcome this divide, we present Mosaic: An efficient, sound and complete verification approach for polynomial real arithmetic properties on piece-wise linear NNs. Mosaic partitions complex verification queries into simple queries and lifts off-the-shelf linear constraint tools to the nonlinear setting in a completeness-preserving manner by combining approximation with exact reasoning for counterexample regions. Our evaluation demonstrates the versatility of VerSAILLE and Mosaic: We prove infinite-time safety on the classical Vertical Airborne Collision Avoidance NNCS verification benchmark for two scenarios while (exhaustively) enumerating counterexample regions in unsafe scenarios. We also show that our approach significantly outperforms State-of-the-Art tools in closed-loop NNV.

Aditi Kabra, Jonathan Laurent, Sagar Bharadwaj et al.

Autonomous cyber-physical systems like robots and self-driving cars could greatly benefit from using formal methods to reason reliably about their control decisions. However, before a problem can be solved it needs to be stated. This requires writing a formal physics model of the cyber-physical system, which is a complex task that traditionally requires human expertise and becomes a bottleneck. This paper experimentally studies whether Large Language Models (LLMs) can automate the formalization process. A 20 problem benchmark suite is designed drawing from undergraduate level physics kinematics problems. In each problem, the LLM is provided with a natural language description of the objects' motion and must produce a model in differential game logic (dGL). The model is (1) syntax checked and iteratively refined based on parser feedback, and (2) semantically evaluated by checking whether symbolically executing the dGL formula recovers the solution to the original physics problem. A success rate of 70% (best over 5 samples) is achieved. We analyze failing cases, identifying directions for future improvement. This provides a first quantitative baseline for LLM-based autoformalization from natural language to a hybrid games logic with continuous dynamics.

David Bayani, Stefan Mitsch

Machine learning is becoming increasingly important to control the behavior of safety and financially critical components in sophisticated environments, where the inability to understand learned components in general, and neural nets in particular, poses serious obstacles to their adoption. Explainability and interpretability methods for learned systems have gained considerable academic attention, but the focus of current approaches on only one aspect of explanation, at a fixed level of abstraction, and limited if any formal guarantees, prevents those explanations from being digestible by the relevant stakeholders (e.g., end users, certification authorities, engineers) with their diverse backgrounds and situation-specific needs. We introduce Fanoos, a framework for combining formal verification techniques, heuristic search, and user interaction to explore explanations at the desired level of granularity and fidelity. We demonstrate the ability of Fanoos to produce and adjust the abstractness of explanations in response to user requests on a learned controller for an inverted double pendulum and on a learned CPU usage model.

Stefan Mitsch, André Platzer

Hybrid systems verification is quite important for developing correct controllers for physical systems, but is also challenging. Verification engineers, thus, need to be empowered with ways of guiding hybrid systems verification while receiving as much help from automation as possible. Due to undecidability, verification tools need sufficient means for intervening during the verification and need to allow verification engineers to provide system design insights. This paper presents the design ideas behind the user interface for the hybrid systems theorem prover KeYmaera X. We discuss how they make it easier to prove hybrid systems as well as help learn how to conduct proofs in the first place. Unsurprisingly, the most difficult user interface challenges come from the desire to integrate automation and human guidance. We also share thoughts how the success of such a user interface design could be evaluated and anecdotal observations about it.