PhD position in artificial intelligence for fisheries resource management

Title

Evidential prediction uncertainty quantification for fish life traits from 3D otolith images

Themes

Artificial Intelligence, Machine Learning, Data Science

Keywords

Dempster-Shafer theory, uncertainty quantification, life traits, marine ecosystems, otolith.

Start date & Duration

September/October 2026, 36 months

Funding

50% IFSEA / 50% Artois University (requested)

Location

The work will be carried out in collaboration between the Laboratory of Computer Engineering and Automation of Artois (LGI2A) in Béthune and the Laboratory of Informatics, Signal and Image of the Opal Coast (LISIC) in Calais.

Subject

Understanding fish life traits (habitat, age, growth, reproduction, longevity, position in the water column, …) is an essential aspect of effective and sustainable management of marine fish stocks. Calcified structures — particularly otoliths, which are the only metabolically inert elements — provide a valuable source of information for this purpose. Notably, their external shape, historically characterized from 2D images and more recently studied in 3D, makes it possible to predict these life traits with high precision. Although 3D images are more informative, they are also more costly to acquire and relatively recent, and therefore less abundant. It is thus necessary to make the best possible use of this rich yet limited source of information in order to obtain the most reliable and accurate predictions. The theory of evidence, also known as Dempster–Shafer theory or the theory of belief functions, is a generalization of the probabilistic framework for reasoning under uncertainty. Its use for uncertainty quantification in predictive modeling is particularly suitable in cases where data are scarce. This PhD project therefore aims to develop predictive methods based on this theory and adapted to current approaches for predicting fish life traits from 3D otolith images. Given the nature of this application, the prediction of ordinal variables will be at the core of the methodological developments.

Supervision

Co‑supervisor: Prof. Emilie Poisson Caillault ( ), University of the Littoral Opal Coast (ULCO), LISIC

Profile Sought

The candidate must hold a master degree or an engineer degree in computer science, applied mathematics or related field. Knowledge in artificial intelligence (machine learning) and/or image processing will be an asset, as well as an awareness of uncertainty management methods. Qualities required to successfully complete a doctoral program such as curiosity, creativity, autonomy, critical thinking, and enthusiasm will be necessary.

Application

Send before April 30, 2026 an e‑mail, with all supervisors of the thesis in copy, containing the following documents in a single pdf file: your CV, a cover letter, your transcripts (with rankings) of the current year (master or equivalent) and previous years, and, optionally, up to two letters of recommendation.