PhD Position: Fairness in Multimodal Machine Learning, Aix-Marseille University

[Hachem Kadri]

[Please feel free to forward this announcement to anyone that you think might be interested]

================================================================

PhD Position: Fairness in Multimodal Machine Learning, Aix-Marseille University
================================================================

A PhD position is open in the Machine Learning Group QARMA at the Computer Science Lab of Aix-Marseille University. The position is available for 3 years starting September 2024 or later. Information about the QARMA Group can be found at https://qarma.lis-lab.fr.

Description: This project forms part of a collaboration between machine learners and neuroscientists in Aix-Marseille University (AMU), with the aim of developing fair multimodal machine learning models. Multimodal learning (or more generally multi-view [1]) considers that the input space is made up of two or more modalities describing examples, where some correlations are supposed to lie across them. Multimodal learning is of interest when one modality of data observation is not sufficient on its own to carry out a learning task with good generalization properties, and/or when each modality carries its own information, sometimes semantically distant from that carried by the other modalities. Research works in multimodal learning have grown over the last decade, within several learning frameworks, from supervised to representation learning, with more than two modalities, etc. For this project, we start from the observation that a modality is itself an observation bias, with its own statistical/topological characterizations and biased shifts of the whole distribution joint over all the views or of a latent space from which all the views would be distinct. This leads to the question of how to define biases in the multimodal setting and how to detect them with respect to the  hypotheses on the joint distributions over modalities. To address this question, the project will develop new theoretical results in the multimodal aspect of fair ML and well-founded and scalable multimodal fair learning algorithms with empirical studies on real datasets from the field of neuro-imaging. A special attention should be paid to graph learning or/and kernel learning.

[1] Paul Pu Liang, Amir Zadeh, and Louis-Philippe Morency (2024) Foundations & Trends in Multimodal Machine Learning: Principles, Challenges, and Open Questions. ACM Comput. Surv. https://doi.org/10.1145/3656580

Supervisors:
* Prof. Cécile Capponi, Machine Learning team (QARMA), Computer Science lab (LIS, CNRS), Aix-Marseille University cecile.capponi(at)lis-lab.fr
* Prof. Hachem Kadri, Machine Learning team (QARMA), Computer Science lab (LIS, CNRS), Aix-Marseille University hachem.kadri(at)lis-lab.fr
This PhD thesis is in collaboration with Benoit Gaüzère from LITIS, Rouen, France.

Selection Criteria: We are looking for a highly motivated candidate with a master degree in machine learning. Experience in the use of machine learning with multimodal data would be strongly appreciated. Strong communication, data presentation and visualization skills. Strong motivation to advance the project by pro-actively developing personal ideas.

Application procedure: All the correspondence regarding this position, including informal inquiry and formal application, should be addressed to Prof. Cécile Capponi and Prof. Hachem Kadri.

Applications must include:
1) A cover letter detailing how you meet the selection criteria for the post;
2) A complete academic CV;
3) Master's transcripts;
3) A sample of scientific output, e.g. a chapter of the thesis;
4) The e-mail contacts of at least two referees who have agreed to provide a reference letter;

Review of the applications will start on the 15th of August at the latest and the position will remain open until a suitable candidate is identified. A first round of interviews is expected to be held no later than the 15th of September and will be held remotely.