We propose a sensitivity-driven framework for constructing Dynamic Bayesian Networks (DBNs) as approximations of Ordinary Differential Equations (ODEs) models while reducing the computational cost of generating training data. The approach uses global sensitivity rankings to identify the most influential direct and indirect dynamical dependencies, which are then used to define reduced sampling supports that capture essential system interactions without exhaustive simulations. The methodology is evaluated on benchmark models of progressively higher dimensionality. A DBN built from a full training dataset serves as a reference and is compared with reduced constructions based on (i) equation sampling using only direct dependencies and (ii) sensitivity-driven supports incorporating both direct and selected indirect influences. This experimental setup allows us to assess whether equation-based sampling alone provides a sufficient approximation of the full model and to quantify the additional benefits of including indirect dynamical effects. Results show that the sensitivity-driven strategy drastically reduces the number of required simulations while maintaining the DBN’s structural and probabilistic fidelity with respect to the original ODE dynamics.

We consider the synthesis problem on timed automata with Büchi objectives, where delay choices made by a controller are subjected to small perturbations. Usually, the controller needs to avoid punctual guards, such as testing the equality of a clock to a constant. In this work, we generalize to a robustness setting that allows for punctual transitions in the automaton to be taken by controller with no perturbation. In order to characterize cycles that resist perturbations in our setting, we introduce a new structural requirement on the reachability relation along an accepting cycle of the automaton. This property is formulated on the region abstraction, and generalizes the existing characterization of winning cycles in the absence of punctual guards. We show that the problem remains within PSPACE despite the presence of punctual guards.

Verification of infinite-state Markov chains is still a challenge despite several fruitful numerical or statistical approaches. For decisive Markov chains, there is a simple numerical algorithm that frames the reachability probability as accurately as required (however with an unknown complexity). On the other hand when applicable, statistical model checking is in most of the cases very efficient. Here we study the relation between these two approaches showing first that decisiveness is a necessary and sufficient condition for almost sure termination of statistical model checking. Afterwards we develop an approach with application to both methods that substitutes to a non decisive Markov chain a decisive Markov chain with the same reachability probability. This approach combines two key ingredients: abstraction and importance sampling (a technique that was formerly used for efficiency). We develop this approach on a generic formalism called layered Markov chain (LMC). Afterwards we perform an empirical study on probabilistic pushdown automata (an instance of LMC) to understand the complexity factors of the statistical and numerical algorithms. To the best of our knowledge, this prototype is the first implementation of the deterministic algorithm for decisive Markov chains and required us to solve several qualitative and numerical issues.

Angluin's L$^*$ algorithm learns the minimal deterministic finite automaton (DFA) of a regular language using membership and equivalence queries. Its probabilistic approximatively correct (PAC) version substitutes an equivalence query by numerous random membership queries to get a high level confidence to the answer. Thus it can be applied to any kind of device and may be viewed as an algorithm for synthesizing an automaton abstracting the behavior of the device based on observations. Here we are interested on how Angluin's PAC learning algorithm behaves for devices which are obtained from a DFA by introducing some noise. More precisely we study whether Angluin's algorithm reduces the noise and produces a DFA closer to the original one than the noisy device. We propose several ways to introduce the noise: (1) the noisy device inverts the classification of words w.r.t. the DFA with a small probability, (2) the noisy device modifies with a small probability the letters of the word before asking its classification w.r.t. the DFA, (3) the noisy device combines the classification of a word w.r.t. the DFA and its classification w.r.t. a counter automaton, and (4) the noisy DFA is obtained by a random process from two DFA such that the language of the first one is included in the second one. Then when a word is accepted (resp. rejected) by the first (resp. second) one, it is also accepted (resp. rejected) and in the remaining cases, it is accepted with probability 0.5. Our main experimental contributions consist in showing that: (1) Angluin's algorithm behaves well whenever the noisy device is produced by a random process, (2) but poorly with a structured noise, and, that (3) is able to eliminate pathological behaviours specified in a regular way. Theoretically, we show that randomness almost surely yields systems with non-recursively enumerable languages.

Sampling timed words out of a timed language described as a timed automaton may seem a simple task: start from the initial state, choose a transition and a delay and repeat until an accepting state is reached. Unfortunately, simple approach based on local, on-the-fly rules produces timed words from distributions that are biased in some unpredictable ways. For this reason, approaches have been developed to guarantee that the sampling follows a more desirable distribution defined over the timed language and not over the automaton. One such distribution is the maximal entropy distribution, whose implementation requires several non-trivial computational steps. In this paper, we present Wordgen which combines those different necessary steps into a lightweight standalone tool. The resulting timed words can be mapped to signals used for model-based testing and falsification of cyber-physical systems thanks to a simple interface with the Breach tool.

Abstract This paper presents a property-directed approach to verifying recurrent neural networks (RNNs). To this end, we learn a deterministic finite automaton as a surrogate model from a given RNN using active automata learning. This model may then be analyzed using model checking as a verification technique. The term property-directed reflects the idea that our procedure is guided and controlled by the given property rather than performing the two steps separately. We show that this not only allows us to discover small counterexamples fast, but also to generalize them by pumping toward faulty flows hinting at the underlying error in the RNN. We also show that our method can be efficiently used for adversarial robustness certification of RNNs.

We develop a novel model-based hardware-in-the-loop (HIL) framework for optimising energy consumption of embedded software controllers. Controller and plant models are specified as networks of parameterised timed input/output automata and translated into executable code. The controller is encoded into the target embedded hardware, which is connected to a power monitor and interacts with the simulation of the plant model. The framework then generates a power consumption model that maps controller transitions to distributions over power measurements, and is used to optimise the timing parameters of the controller, without compromising a given safety requirement. The novelty of our approach is that we measure the real power consumption of the controller and use thus obtained data for energy optimisation. We employ timed Petri nets as an intermediate representation of the executable specification, which facilitates efficient code generation and fast simulations. Our framework uniquely combines the advantages of rigorous specifications with accurate power measurements and methods for online model estimation, thus enabling automated design of correct and energy-efficient controllers.

The Hybrid Automata Stochastic Logic (HASL) has been recently defined as a flexible way to express classical performance measures as well as more complex, path-based ones (generically called "HASL formulas"). The considered paths are executions of Generalized Stochastic Petri Nets (GSPN), which are an extension of the basic Petri net formalism to define discrete event stochastic processes. The computation of the HASL formulas for a GSPN model is demanded to the COSMOS tool, that applies simulation techniques to the formula computation. Stochastic Symmetric Nets (SSN) are a high level Petri net formalism, of the colored type, in which tokens can have an identity, and it is well known that colored Petri nets allow one to describe systems in a more compact and parametric form than basic (uncolored) Petri nets. In this paper we propose to extend HASL and COSMOS to support colors, so that performance formulas for SSN can be easily defined and evaluated. This requires a new definition of the logic, to ensure that colors are taken into account in a correct and useful manner, and a significant extension of the COSMOS tool.