Sylvain Combettes' projects

Oct. 2019 – Feb. 2020 (5 months)
▸ Context: For my final-year project at Mines Nancy (one day per week), I worked on a project for a research departement of the CNRS, the largest governmental research organisation in France.
▸ Topic: Comparison of empirical probability distributions. Application to the Choquet integral with stochastic inputs.
▸ Methods: Integral probability metrics (e.g. Kantorovich metric), f-divergences (e.g. Kullback-Leibler).
▸ Programming: Python.
▸ Result: We empirically show that a new method for simulating the Choquet integral is "correct".
▸ Links: [GitHub repository] [full 62 pages report] [slides]

June 2019 – Sep. 2019 (3 months)
▸ Context: As part of my penultimate-year at Mines Nancy, I did a 3-month research internship at Servier, the second largest pharmaceutical company in France. In 2018, Servier had a €4.2 billion turnover, operated in 149 countries and had more than 22,000 employees.
▸ Topic: Generating fictitious realistic patient data in order to boost the prediction score [synthesis, dataset augmentation].
▸ Method: Combining GANs (generative adversarial networks) with autoencoders [implicit density estimation].
▸ Programming: Python.
▸ Result: The prediction score can be increased by more than 5% on binary values.
▸ Links: [GitHub repository] [5 pages synthetic report] [full 62 pages report] [slides]

Oct. 2018 – June 2019 (9 months)
▸ Context: For my penultimate-year project at Mines Nancy (half a day per week), I did research for the French company Saint-Gobain, the European or worldwide leader in all of its businesses (mainly construction materials). In 2018, Saint-Gobain had a €41.8 billion turnover, operated in 67 countries and had more than 180,000 employees.
▸ Topic: Detection of sensor failure in a production line [anomaly detection].
▸ Methods: Principal component analysis (PCA) and kernel principal component analysis (KPCA).
▸ Programming: MATLAB.
▸ Result: The algorithm can detect 100% of the failure days observed by Saint-Gobain.
▸ Links: [GitHub repository] [report incoming]