PhD Position F/M Ph.D. student: Nonsmooth optimization for machine learning

Contract type : Fixed-term contract

Level of qualifications required : Graduate degree or equivalent

Fonction : PhD Position

About the research centre or Inria department

The Centre Inria de l’Université de Grenoble groups together almost 450 people in 26 research teams and 9 research support departments.

Staff is present on three campuses in Grenoble, in close collaboration with other research and higher education institutions (Université Grenoble Alpes, CNRS, CEA, INRAE, …), but also with key economic players in the area.

The Centre Inria de l’Université Grenoble Alpes is active in the fields of high-performance computing, verification and embedded systems, modeling of the environment at multiple levels, and data science and artificial intelligence. The center is a top-level scientific institute with an extensive network of international collaborations in Europe and the rest of the world.

Context

The Ph.D. research will be carried out between the GHOST team at Inria and the DAO team of the LJK laboratory.
The candidate will be funded by the research chair on Structured Optimization and Learning of the MIAI cluster.

The goal will be to develop and analyze new optimization algorithms that are well-suited to the nonsmooth nonconvex problemsthat are common in machine learning. While modern deep architectures often include nonsmooth layers, many currently popular methods are mostly studied and understood through the lens of smooth optimization.
We will therefore strive to establish theoretical guarantees on algorithms close to those used by machine learning practitioners in order to close this theory-practice gap.

Assignment

The recruited candidate will be co-supervised by Mathieu Besançon (Inria, LIG), Quoc-Tung Le (UGA, LJK) and Jérôme Malick (CNRS, LJK).

To understand the topic:

Some algorithms studied in the smooth case are currently ill-understood when applied to nonsmooth instances. One of the research questions will focus on the definition of a problem class close to learning models and on which nonsmooth algorithms can be analyzed.

The candidate can read the recent literature on adaptive algorithms such as AdaGrad for relevant families of nonsmooth methods that will be considered.

Collaboration: Some collaborations with the Zuse Institute Berlin and the Institut Mathématique de Toulouse will be possible during the Ph.D.

Main activities

- Formalizing necessary properties for the convergence of optimization algorithms on nonsmooth functions.
- Developing new algorithms that are well-suited to problem classes close to modern learning architectures.
- Implementing and evaluating the computational performance of the proposed methods on reference benchmark problems.

Complementary activities:

- Developing open-source software implementations of optimization methods
- Writing reports and research articles

Skills

Mathematical methods for optimization: analysis, convex analysis, linear algebra, analysis of algorithms, complexity.
Languages: written and spoken English, French is optional.

Benefits package

  • Subsidized meals
  • Partial reimbursement of public transport costs
  • Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
  • Possibility of teleworking (90 days / year) and flexible organization of working hours
  • Social, cultural and sports events and activities
  • Access to vocational training
  • Social security coverage under conditions

Remuneration

2200 euros gross salary /month