Post-doctorat en interprétabilité mécaniste et diversité des représentations

About This Opportunity

We are representing a leading French research institution in their search for a postdoctoral researcher specializing in mechanistic interpretability and representation diversity. This 18-month position is based in Brest, Brittany, and focuses on advancing our understanding of how diversity—in both data and internal representations—impacts machine learning systems, particularly in low-annotation regimes.

Our client is a prominent academic research organization working at the forefront of artificial intelligence, with particular expertise in foundation models, signal processing, and applications in healthcare. The research team has recently developed groundbreaking work in EEG foundation models and discrete diffusion language models.

Responsibilities

• Train and analyze sparse autoencoders on foundation model activations to extract interpretable features and descriptors
• Develop and evaluate diversity criteria in representation spaces using mechanistic interpretability tools
• Construct similarity kernels between descriptors and assess coverage using determinantal point processes
• Investigate the relationship between training data diversity and the diversity of internal mechanisms that models acquire
• Compare features extracted by sparse autoencoders across models trained on different data regimes and quantify their overlap
• Evaluate the stability of discovered features across training runs to distinguish reproducible mechanisms from optimization artifacts
• Measure the effect of diversity criteria on downstream tasks in low-annotation settings
• Publish findings in top-tier international conferences and journals
• Release code and experimental protocols to ensure reproducibility
• Collaborate directly with multiple doctoral students working on related aspects of the project

What We're Looking For

Essential qualifications:

• PhD obtained within 3 years prior to start date in machine learning, computer science, signal processing, or applied mathematics
• Strong mastery of deep learning and its mathematical foundations
• Proficiency in Python and a deep learning framework (PyTorch preferred)
• Hands-on experience working with transformer architectures and instrumenting their internal representations
• Autonomy in conducting GPU-based experiments, including on shared computing infrastructure
• Track record of publications in international ML conferences or journals
• Excellent scientific English, both written and spoken
• Scientific autonomy and enthusiasm for open-ended research questions
• Experimental rigor with attention to reproducibility
• Strong collaborative skills and ability to work with doctoral students
• Strong writing abilities and capacity to present results to non-specialist audiences

Highly desirable:

• Knowledge of mechanistic interpretability: sparse autoencoders, linear probes, circuit analysis
• Familiarity with determinantal point processes or sampling methods and random linear algebra
• Interest in discrete diffusion language models
• Experience with multi-GPU distributed computing and SLURM-type schedulers

Note: French language skills are not required—the research team operates in English when any member is not francophone.

How We Work With You

When you apply through Get A Job.ai, our recruiting team will carefully review your background against the requirements of this confidential research position. Qualified candidates will be screened by one of our recruiters who specializes in academic and research placements. We will then present your profile to our client for consideration.

Please submit your application exclusively through the Get A Job.ai platform. Do not attempt to contact the hiring institution directly, as this may disqualify your candidacy.

Pay

Pay via Get A Job.ai: €35,800 gross annual (subject to profile and experience; reduced employee contributions in the public sector)

Position details:

• Contract type: 18-month fixed-term contract under public law (renewable)
• Location: Brest, Brittany, France (on-site with partial remote work possible)
• Start date: January 1, 2027
• Application deadline: October 10, 2026

Additional benefits include: on-site dining facilities, sports and leisure offerings, public transport reimbursement, sustainable mobility allowance, family supplement, comprehensive social benefits, generous leave allowances

Get A Job.ai is an equal opportunity recruiter. We welcome applications from all qualified candidates and can arrange accommodations for candidates with disabilities upon request.