Supervisors: Eugene Asarin and Peter Habermehl
Location: IRIF (Paris)
Context: Franco-Japanese project CyPhAI on a related theme started in November 2020, PhD thesis available.
Hybrid automata combine discrete behavior (states and transitions) with continuous one (differential equations). They are extensively used for modelling, analysis and verification of cyber-physical systems (such as autonomous cars, medical equipment etc.)
Given a set of data (examples, and possibly counterexamples, of admissible behaviors of a system, involving discrete events and numeric values), we aim to build (to learn) a hybrid automaton explaining this data.
The intern will start with bibliographic study on automata-based methods of learning finite-state and timed automata. She or he will next develop and prove learning procedures for simple subclasses of hybrid automata, implement them, and test them on real or synthetic data. Finally, she will participate in a brainstorming on neural networks in learning hybrid automata.
We believe that the subject is very promising and could be extended to a good PhD thesis.
Basic knowledge of automata, algorithmics, differential equations, taste for fundamental informatics and good programming skills are mandatory. The ideal candidate should know basics of timed and hybrid automata (e.g. course 2-8-2 of MPRI), of machine learning and artificial intelligence would be welcome.