Summary
The Centre Inria de l’Université Grenoble Alpes is offering a 3-year PhD position funded by the MIAI cluster’s research chair on Structured Optimization and Learning. The successful candidate will conduct research jointly between the GHOST team at Inria and the DAO team at the LJK laboratory, focusing on developing and analyzing new optimization algorithms tailored for nonsmooth, nonconvex problems in modern machine learning architectures.
PhD Position F/M: Nonsmooth Optimization for Machine Learning, Centre Inria de l’Université, France
At a Glance
| Feature | Details |
| Offer Number | #2026-10342 |
| Contract Type | Fixed-term (3 years) |
| Starting Date | October 1, 2026 |
| Remuneration | €2,200 gross salary / month |
| Benefits | Subsidized meals, transport reimbursement, 7 weeks annual leave + 10 RTT days, up to 90 days/year teleworking |
Subject Area
Optimization, machine learning, statistical methods, and scientific computing (BAP E).
Location
Saint-Martin-d’Hères (Centre Inria de l’Université Grenoble Alpes), France.
Eligibility
- Academic Level: Graduate degree or equivalent.
- Technical Skills: Strong foundation in mathematical methods for optimization, including mathematical analysis, convex analysis, linear algebra, analysis of algorithms, and complexity.
- Languages: Proficiency in written and spoken English is required; French is optional.
- Traits: Mathematical rigor and scientific curiosity.
Scholarship Description
While modern deep learning architectures frequently utilize nonsmooth layers, popular optimization methods are primarily understood through the lens of smooth optimization. This PhD aims to bridge the theory-practice gap by analyzing algorithms close to those actually used by machine learning practitioners.
Main Activities:
- Formalizing necessary properties for the convergence of optimization algorithms on nonsmooth functions.
- Developing new algorithms suited to problem classes close to modern learning architectures (such as studying adaptive algorithms like AdaGrad for relevant families of nonsmooth methods).
- Implementing and evaluating the computational performance of proposed methods on reference benchmark problems.
- Developing open-source software implementations of optimization methods.
- Writing reports and research articles.
How to Apply
Applications must be submitted online via the Inria website. You must enter your e-mail address to save your application to Inria. Please note that applications sent through other channels are not guaranteed to be processed. Additionally, as this position may be located in a restricted area (ZRR) for national security, an authorization from the Ministry will be required.
Last Date
August 26, 2026
