Le 21/10/2019 par Webmaster ROADEF :
The interdisciplinary institute in artificial intelligence of Toulouse, named the Artificial and Natural Intelligence Toulouse Institute (ANITI), is one of four institutes spearheading research on AI in France. Part of a 24 chairs' program funded by ANITI, Thomas Schiex's chair is on Pushing the computational frontiers of reasoning with logic, probabilities and preferences. The chair and co-chair Simon de Givry, working in a bioinformatic team at INRA Toulouse, are seeking a postdoctoral fellow. The position is available immediately. The project will be funded on a contract for at most four years with net salary of 2600€ or more per month with some teaching (64 hours per year on average).
Constraint programming (CP) is an AI Automated Reasoning technology with tight connections with propositional logic. It offers a problem modeling and solving framework where the set of solutions of a complex (NP-hard) problem is described by discrete variables, connected by constraints (simple Boolean functions). Together with propositional satisfiability, it is one of the automated reasoning approaches of AI, where problems are solved exactly to provide rigorous solutions to hardware or software testing and verification, system configuration, scheduling or planning problems.
Discrete Stochastic Graphical Models (GMs) define a Machine Learning technology where a probability mass function is described by discrete variables, connected by potentials (simple numerical functions). GMs can be learned from data and the NP-hard problem of identifying a Maximum a Posteriori (MAP) labelling is often solved approximately to tackle several problems in Image and Natural Language Processing, among others.
The Cost Function Network framework with its associated C++ open source award-winning solver toulbar2, developed in our team, combine the ideas of Constraint Programming and Stochastic Graphical Models. By solving the so-called Weighted Constraint Satisfaction problem, toulbar2 is capable of simultaneously reasoning on logical information described as Boolean functions and gradual, possibly Machine Learned, information described as local numerical functions.
To process the available information, the solver relies on a guaranteed hybrid branch and bound algorithm. In this algorithm, pruning follows from a variety of mechanisms that can either simplify the problem at hand, provide primal solutions (using local search, rounding or incomplete tree-search), or provide dual solutions and lower bounds. Parallel solving offers new opportunities to organize these various mechanisms differently in time, to exploit problem decompositions, to apply stronger primal/dual reasoning, and to use Machine Learning to guide search or decide which mechanism to activate based on the current solving and/or a collection of instances of the same problem.
Experiments will be performed on large collections of real problem instances, many of which are not known to be currently solvable. This includes the possible application of toulbar2 onto current exciting problems in Computational Protein Design (CPD), in collaboration with molecular modellers and biochemists, and in the context of the ongoing development of a dedicated CPD software.
The position is specifically open to highly creative researchers that may quickly want to develop and explore their own ideas. As such, we expect that the PostDoc will be increasingly capable of injecting their own ideas in the project, in interaction with all the members of the project team as well as external collaborators, and contribute to the supervision of PhD students.
Candidate profile
The PostDoc is at the intersection of CP, SAT, integer programming, metaheuristics, and distributed computing. The ideal candidate should therefore be familiar with CP or SAT algorithms. He or she may also benefit from background knowledge in the weighted variants of SAT/CP, in Integer Linear Programming, or in Stochastic Graphical Models processing. Some experience in the design and implementation of multi-threaded/distributed code is a nice plus. Good programming abilities (in C++ ideally) will be required. Additional knowledge in bioinformatics, biochemistry, and molecular modelling would be a plus in the context of CPD applications.
How to apply
Please email your detailed CV, a motivation letter, and transcripts of bachelor's degree and PhD in Computer Science to simon.de-givry@inra.fr and thomas.schiex@inra.fr. Samples of published research by the candidate and reference letters will be a plus.
APPLICATION DEADLINE FOR FULL CONSIDERATION: December 1, 2019.
https://mia.toulouse.inra.fr/images/2/26/PostDocANITI.pdf
https://en.univ-toulouse.fr/aniti
http://www7.inra.fr/mia/T/schiex/
http://www7.inra.fr/mia/T/degivry/
http://www7.inra.fr/mia/T/toulbar2/
