Formal methods for neural network-based systems

Point of contact

  • Alexandre Chapoutot (ENSTA Paris)
  • Franck Delaplace (IBISC, UEVE)


Neural networks are an important class of algorithms used in autonomous systems or robots for perception or control. The size and the complexity of such neural networks represent a new challenge to apply formal methods on closed-loop systems. As a side effect, an new research direction is to build safe-by-construction controller based neural networks. Dealing with neural networks component involve also the design of formal specification which is a new challenge.


Formal methods, neural networks, closed-loop systems, controller

Researchers involved or interested

A few references

  • Reachability Analysis for Neural Feedback Systems using Regressive Polynomial Rule Inference by Souradeep Dutta, Xin Chen and Sriram Sankaranarayanan. HSCC 2019.
  • Verisig: verifying safety properties of hybrid systems with neural network controllers by Radoslav Ivanov, James Weimer, Rajeev Alur, George Pappas and Insup Lee. HSCC 2019.
  • Probabilistic Guarantees for Safe Deep Reinforcement Learning by Edoardo Bacci, David Parker. FORMATS 2020.

Related master programs