Funding

Details; France; Sur place; EUR 30 000 - 42 000; Plein temps

Résumé du poste

A leading research institution in France seeks a Recognised Researcher (R2) to develop a rigorous regularization theory for hybrid deep-learning algorithms. The candidate must have a strong background in applied mathematics and good programming skills. This post-doc aims to tackle significant questions related to inverse problems and deep learning applications, with the position starting January 1, 2026.

Qualifications

• Strong background in applied mathematics, particularly in optimization and data science.
• Good programming skills.
• Good communication skills in English, both written and oral.

Responsabilités

• Develop a rigorous regularization theory of hybrid deep-learning algorithms.
• Tackle questions regarding stability to noise and convergence rates.

Connaissances

Applied mathematics

Optimization

Data science

Programming skills

Communication in English

Formation

PhD or equivalent

Description du poste

Recognised Researcher (R2) (PhD holders or equivalent who are not yet fully independent)

About

Context
The context of this post‑doc is within the ”IA for science” project funded by the University
of Caen CAESAR ”Excellence sous toutes ses formes” program. It involves a large collaboration
between mathematicians, computer scientists and physicists. It is concerned with designing a rigorous
regularization theory of machine learning‑based approaches, in particular those on neural
networks, that have become ubiquitous for solving a variety ionic inverse problems in various fields
including physics.

Project overview
Inverse problems (i.e., recovering an object from its indirect and noisy measurements) have
historically been solved by formulating them as a variational problem, where the objective functional
to be minimized has a composite structure, and involving a cost function derived from a physics‑based
forward model. Such model‑based methods provide explainable results in a computationally‑efficient
way and benefit from a wealth of recovery theoretical results. Yet their application remains
limited by the physical model they rely on and the hand‑crafted nature of the regularizer. On the
other hand, in the last decade, the rise of deep learning has demonstrated that these limits may
eventually be overcome by using either pure data‑driven or hybrid methods that combine the
best of both worlds. These include PnP methods, unrolling and generative models. While these
methods have become popular and heavily developed in the last years, leading to qualitatively impressive results, they are lacking a thorough theoretical understanding, in particular of their recovery guarantees. In this post‑doc, our goal is to partly close that gap for some of the above methods.

Objectives and scientific program

The objective of this post‑doc is to develop a rigorous regularization theory of state‑of‑the‑art
hybrid deep‑learning based algorithms for solving inverse problems. Several natural questions not yet answered in the literature will be tackled in this post‑doc:

• Under what conditions and in what sense would we expect such methods to recover a ground truth object from its indirect measurements in the absence of noise?
• Under what conditions and in what sense would we ensure stability to noise, learning, etc., and convergence as the different errors involved vanish?
• At which rate does such convergence take place?
• What is the influence of the learning step on the overall inversion process guarantees?

The candidate must have a strong background in applied mathematics, in particular in optimization and data science. Background in imaging and computer vision is definitely a plus. The candidate should also have good programming skills, and good communication skills in English, both written and oral.

Languages

ENGLISH

Level

Good

Research Field

Mathematics, Computer science

Years of Research Experience

1 - 4

Additional Information

Selection process

To apply:

Please submit the following documents via email:

• A CV including a list of publications.
• A cover letter outlining your research statement, background, and motivation for applying.
• A list of references to be contacted.

The position is to be filled January 1 2026. Application until 13/11/2025.

Website for additional job details

Work Location(s)

Number of offers available

1

Company/Institute

The responsibility for the funding offers published on this website, including the funding description, lies entirely with the publishing institutions. The application is handled uniquely by the employer, who is also fully responsible for the recruitment and selection processes.