How we work

How we work

The CentraleSupélec - Lusis Research Chair is a long-term industrial-academic partnership conducting applied AI research on three axes: fraud detection, automated trading, and global health. The sections below describe how the chair operates day to day.

Research at three levels

We host and supervise researchers at three levels:

  • PhD students. Multi-year doctoral projects supervised by chair faculty, each anchored on one of the research axes. See current and past doctoral candidates on the PhDs page.
  • Graduate projects. Last-year CentraleSupélec students conduct a year-long research project in groups of 2 to 5, on a chair-defined topic. Projects from past cohorts are listed on the Graduate projects page.
  • Internships. Master-level research internships hosted within the chair, typically over a 3–6 month period. Past internships will be listed once that programme begins running.

Each level feeds into the next: graduate projects let us probe new directions cheaply, productive lines mature into PhD subjects, and PhD students often co-supervise the next graduate cohort.

Working with LUSIS

LUSIS is the chair’s industrial partner. Concretely the collaboration takes three forms:

  • Regular working meetings. Researchers and LUSIS engineers meet on a recurring schedule to align research directions with practical needs, review preliminary results, and re-formulate problems when production reality contradicts a clean academic statement.
  • Real-world data. LUSIS and its banking partners provide access to anonymised transaction datasets at production scale — hundreds of millions of records — so methods are evaluated on data that reflects industrial conditions (heavy class imbalance, concept drift, latency budgets) rather than curated academic benchmarks.
  • Joint output. Where appropriate, results are published jointly: LUSIS engineers co-author scientific articles, co-present at conferences (recent venues include NeurIPS, ICLR, ICML, CIKM, ESANN, ILB), and contribute to the chair’s research code.

Computing infrastructure

Training and evaluating modern machine-learning models at industrial scale requires significant compute. The chair relies on two shared academic HPC facilities for its experiments:

  • Mésocentre Paris-Saclay — the regional research-computing facility operated jointly by Université Paris-Saclay, CentraleSupélec, and ENS Paris-Saclay, with support from CNRS and the Île-de-France region. Access is open to laboratories within the Paris-Saclay perimeter. The chair uses it for medium-scale CPU and GPU workloads: prototyping, hyperparameter sweeps, and dataset preprocessing.
  • Jean Zay — the French national converged supercomputer operated by IDRIS (the CNRS national HPC centre), purpose-built for AI workloads alongside traditional scientific computing. Its large GPU partitions are used for the chair’s heavier training runs — deep anomaly-detection architectures on full-scale LUSIS transaction datasets, self-supervised models for tabular data, and similar. Access is granted via the national GENCI allocation process.

Getting involved

Open positions — PhD topics, graduate project openings, and internships — are announced on the Jobs page. For other enquiries, please contact us.