IMT Atlantique
Just Landed
Posted Yesterday
Post-doctoral researcher in mechanistic interpretability and diversity of representations IMT Atlantique in France
Degree Level
Postdoc
Field of study
Computer Science
Funding
Available
Deadline
Oct 10, 2026
Country
France
University
IMT Atlantique
How do I apply for this?
Sign in for free to reveal details, requirements, and source links.
More information can be found here
Keywords
Suggested positions
About this position
IMT Atlantique is advertising a post-doctoral researcher position in mechanistic interpretability and diversity of representations, based in Brest, France, within the BRAIN team (BRoader Artificial INtelligence) of Lab-STICC (UMRCNRS 6285).
The position is part of the ENDIVE project, which studies how diversity can help machine learning when annotation budget is limited. The project investigates diversity at two levels: the diversity of training data and the diversity of internal representations acquired by models. This postdoc focuses on the second level, using tools from mechanistic interpretability.
The research will involve training and analysing sparse autoencoders on foundation-model activations, building similarity kernels over learned features, and using determinantal point processes to test whether diverse subsets can cover representations more efficiently. The work also includes studying how different data diets affect the mechanisms learned by models, comparing stability across training runs, and relating these findings to curation strategies for discrete diffusion language models.
The role is research-intensive and aims to contribute to scientific publications, code release, experimental reproducibility, and project dissemination. The successful candidate will also interact with and help supervise interns in the project environment.
Eligibility highlights: a PhD obtained less than 3 years before the hire date in machine learning, computer science, signal processing, or applied mathematics; strong deep learning and Python/PyTorch skills; experience with transformer architectures and internal representations; and the ability to run GPU experiments independently. Publications in the field and strong scientific English are expected. Familiarity with mechanistic interpretability, DPPs, sampling methods, discrete diffusion, or SLURM-like multi-GPU computing is welcome but not mandatory. French is not required; English is the working language of the team.
Application deadline: 10 October 2026. Start date: 1 January 2027. The contract is full-time and temporary.
Funding details
Available
What's required
Applicants must hold a PhD obtained less than 3 years before the hire date in machine learning, computer science, signal processing, or applied mathematics. Required skills include a solid command of deep learning and its mathematical foundations, strong Python programming skills with preferably PyTorch, experience with transformer architectures and instrumenting internal representations, and the ability to run GPU experiments on shared computing infrastructure. Candidates should have publications in international conferences or journals and a very good level of scientific English, written and spoken. Familiarity with mechanistic interpretability, determinantal point processes or sampling methods, discrete diffusion language models, or multi-GPU computing on a SLURM-like scheduler is appreciated but not required. The role also requires scientific autonomy, experimental rigor, attention to reproducibility, teamwork, and good writing and presentation skills. French is not required; English is the working language.
How to apply
Apply through the IMT Atlantique Recruitee webpage using the provided application link. Submit before 2026-10-10. Interviews are planned in real time.
Ask ApplyKite AI
How do I apply for this?
Sign in for free to reveal details, requirements, and source links.