Abstract:Automatic verification of concurrent programs faces state explosion due to the exponential possible interleavings of its sequential components coupled with large or infinite state spaces. An alternative is deductive verification, where given a candidate invariant, we establish inductive invariance and show that any state satisfying the invariant is also safe. However, learning (inductive) program invariants is difficult. To this end, we propose a data-driven procedure to synthesize program invariants, where it is assumed that the program invariant is an expression that characterizes a (hopefully tight) over-approximation of the reachable program states. The main ideas of our approach are: (1) We treat a candidate invariant as a classifier separating states observed in (sampled) program traces from those speculated to be unreachable. (2) We develop an enumerative, template-free approach to learn such classifiers from positive and negative examples. At its core, our enumerative approach employs decision trees to generate expressions that do not over-fit to the observed states (and thus generalize). (3) We employ a runtime framework to monitor program executions that may refute the candidate invariant; every refutation triggers a revision of the candidate invariant. Our runtime framework can be viewed as an instance of statistical model checking, which gives us probabilistic guarantees on the candidate invariant. We also show that such in some cases, our counterexample-guided inductive synthesis approach converges (in probability) to an overapproximation of the reachable set of states. Our experimental results show that our framework excels in learning useful invariants using only a fraction of the set of reachable states for a wide variety of concurrent programs.

Scenario-Based Set Invariance Verification for Black-Box Nonlinear Systems.

Data-driven computation of invariant sets of discrete time-invariant black-box systems

Data-driven Invariant Subspace Identification for Black-Box Switched Linear Systems.

A Framework for Safe Probabilistic Invariance Verification of Stochastic Dynamical Systems

A Physics-Informed Scenario Approach with Data Mitigation for Safety Verification of Nonlinear Systems

Safe Probabilistic Invariance Verification for Stochastic Discrete-Time Dynamical Systems

Safety Verification of Nonlinear Hybrid Systems Based on Invariant Clusters.

A Formal Characterization of Black-Box System Safety Performance with Scenario Sampling

Detection of linearly dependent and inconsistent equations in process system models

Data-driven Control of unknown Switched Linear Systems using Scenario Optimization

Data-Driven Scenario Optimization for Automated Controller Tuning with Probabilistic Performance Guarantees

Forward Invariance in Neural Network Controlled Systems

Scenario based optimisation over uncertain system identification models

Formal Verification of Unknown Stochastic Systems via Non-parametric Estimation

Statistical Verification of Autonomous Systems using Surrogate Models and Conformal Inference

Scenario Reduction with Guarantees for Stochastic Optimal Control of Linear Systems

Invariant Clusters for Hybrid Systems

Structural Invariants for Parametric Verification of Systems with Almost Linear Architectures

Robust Invariant Sets Computation for Discrete-Time Perturbed Nonlinear Systems

Data-Driven Controlled Invariant Sets for Gaussian Process State Space Models

Data-Driven Template-Free Invariant Generation