PhD Proposal: A Unified Approach for Learning Temporal Behaviors

The problem of identifying patterns in temporal behaviors (sequences, signals) is treated in different disciplines using different methods. In theoretical computer science, for example, there are algorithms for learning regular languages from examples, by constructing the minimal accepting automaton. In the "deep learning" literature, there are various algorithms that use recurrent neural networks (RNN) for this purpose. Likewise there are techniques coming from statistics, control and signal processing. The goal of this thesis is to gain a better understanding of the relative respective merits of those techniques, with the goal of developing powerful learning algorithms for data mining of cyber-physical systems, focusing on specific issues that are considered important like multi-dimensionality, the lack of negative examples, etc.

The potential funding for this position is to be gained in a competition of the local doctoral school. Please apply to this position only if you have an excellent record from your undergraduate/master studies. Please send me a mail including a CV and motivation letter with "PhD position" in the title before 10/3/18.

Further reading:

O. Maler, I.E. Mens, A Generic Algorithm for Learning Symbolic Automata from Membrship Queries, Models, Algorithms, Logics and Tools 2017 New

E. Asarin, A. Donze, O. Maler, D. Nickovic, Parametric Identification of Temporal Properties, RV 2011 Slides