Welcome!

This is the website of Pierre Talbot. I'm a research scientist at the University of Luxembourg. I'm working on new theories of constraint reasoning and parallel programming. I'm also the co-director of the Master in High Performance Computing.

News

  • 🎉10 July 2024: Hedieh Haddad got her paper Comparison of Hyperparameter Optimization Methods for Selecting Search Strategy of Constraint Programming Solvers accepted to PTHG-24.
  • 17 June 2024: We organize the new Abstract Interpretation Workshop 2024, feel free to join :-)
  • 26 April 2024: Our proposal with Angelica Rings A Musical Game to Understand the Challenge of Collaboration in Computing for the Researchers' Days (28-30 November 2024) has been accepted.
  • 23 April 2024: I gave a seminar to the parallel and combinatorial optimization group (PCOG) on an introduction to abstract interpretation, slides.

Abstract Constraint Reasoning

Can abstract interpretation be the backbone theory to unify constraint reasoning approaches?
  • Goal I: Combine constraint solvers (by reduced products) to more efficiently solve combinatorial problems.
  • Goal II: Generalize reasoning procedures (e.g., propagation, multiobjective algorithms, clause learning) to monotone functions working over any abstract domains.
We are working on a new theory of combinatorial optimization called abstract constraint reasoning. It relies on lattice theory and abstract interpretation to unify the subfields of combinatorial optimization.

Highlights:
  • The lattice land project is a collection of C++ libraries implementing abstract domains such as intervals, octagons and new ones such as propagator completion.
  • Our TPLP journal paper (2020) introduces the propagator completion and new a product of abstract domains.

Parallel Lattice Programming

Can lattice theory be the backbone of a safe model of parallel programming?
  • Goal I: Make parallel programs correct-by-construction.
  • Goal II: Take advantage of specialized hardware (e.g., GPUs, FPGAs, quantum architectures)
The foundation of this language is the same than for abstract constraint reasoning: lattice theory. In short: data are lattices, programs are monotone functions and the execution is the computation of a fixpoint. Our primary focus and application is to accelerate the algorithms developed for abstract constraint reasoning on parallel architectures.

Highlights:
  • Turbo is an abstract constraint solver fully executing on the GPU.
  • cuda-battery provides C++ data structures working on both the CPU and GPU (CUDA).
  • Our AAAI2022 paper describes the foundation of this parallel model of computation.
  • FNR CORE Grant COMOC 2022-2025 to explore this strand of research (PI: P. Talbot, 384k€).

Team

I have the pleasure to co-supervise and work with several Ph.D. students and a postdoc.

Pierre Talbot
Thibault Falque
Hedieh Haddad
Manuel Combarro

Yi-Nung Tsao
Tobias Fischerbach

  • Thibault Falque, postdoctoral researcher on the project COMOC, 2024-2025.
  • Hedieh Haddad, Ph.D. candidate, Hyperparameter Optimization of Constraint Solver, 2023-2025.
  • Manuel Combarro, Ph.D. candidate, Multiobjective Constraint Programming, 2023-2025.
  • Yi-Nung Tsao, Ph.D. candidate, Verification of Neural Networks by Abstract Interpretation, 2023-2026.
  • Tobias Fischerbach, Ph.D. candidate, Optimization of Quantum Circuits, 2023-2026.

Previously...

A lattice-based approach for GPU programming (Postdoc 2020-2023)
Abstract domains for constraint programming (Postdoc 2018-2019)
Spacetime Programming: A Synchronous Language for Constraint Search (Ph.D. 2014-2018)
  • bonsai is a language implementing the spacetime paradigm for Java.
  • pcp is a library for constraint solving written in the language Rust.
  • oak is a PEG parser investigating the notion of AST inference.