Phd Position F/m Private and Fair Machine Learning

Le descriptif de l’offre ci-dessous est en Anglais_

**Type de contrat**: CDD

**Contrat renouvelable**: Oui

**Niveau de diplôme exigé**: Bac + 5 ou équivalent

**Fonction**: Doctorant

**A propos du centre ou de la direction fonctionnelle**:
The Centre Inria de l’Université de Grenoble groups together almost 600 people in 24 research teams and 9 research support departments.

Staff is present on three campuses in Grenoble, in close collaboration with other research and higher education institutions (Université Grenoble Alpes, CNRS, CEA, INRAE,), but also with key economic players in the area.

The Centre Inria de l’Université Grenoble Alpes is active in the fields of high-performance computing, verification and embedded systems, modeling of the environment at multiple levels, and data science and artificial intelligence. The center is a top-level scientific institute with an extensive network of international collaborations in Europe and the rest of the world.

**Contexte et atouts du poste**:
**Context.** This PhD thesis is part of the ANR JCJC project AI-PULSE (Aligning Privacy, Utility, and Fairness for Responsible AI), coordinated by Héber H. Arcolezi. AI-PULSE, which started in March 2025, aims to design machine learning models that are both differentially private and fairness-aware.

The thesis will be conducted under a co-tutelle agreement between Inria Grenoble (France) and ÉTS Montreal (Canada), leveraging the complementary strengths of both institutions.

The envisioned plan is for the recruited PhD student to spend approximately two years at Inria Grenoble (Privatics team), followed by two years at ÉTS Montreal. This structure will provide a rich and balanced international training environment, enabling the student to benefit from diverse expertise, ecosystems, and research cultures.

The thesis will be co-supervised by:

- Héber H. Arcolezi, Researcher at Inria and in-coming Assistant Professor at ÉTS Montreal (February 2026).
- Claude Castelluccia, Researcher at Inria Grenoble and CNIL Commissioner.
- Ulrich Aïvodji, Assistant Professor at ÉTS Montreal.

The PhD project will contribute to the core objectives of AI-PULSE, with a particular focus on advancing methodologies that jointly address privacy and fairness in machine learning under local differential privacy. The student will also benefit from the broader international collaborations and mobility opportunities enabled by the co-tutelle agreement, further strengthening the international dimension of their training and research.

**Mission confiée**:
**Assignment.** Modern _Machine Learning_ (ML) systems increasingly drive automated decision-making across sectors like healthcare, finance, and public policy. While these systems can offer remarkable benefits, they also raise critical concerns regarding individual privacy and algorithmic fairness.
- Differential Privacy_ (DP) [1] has emerged as a gold-standard privacy notion for balancing the privacy-utility trade-off in data analytics. However, the central (server-side) approach to DP requires trust in a third party holding the raw data. Hence, there is growing interest in _Local Differential Privacy_ (LDP) [2], which performs data obfuscation directly at the user’s side, removing the need to trust a central server with unprotected personal information.

Meanwhile, _fairness in ML_ [3] typically involves mitigating disparate impact among subgroups defined by sensitive attributes (e.g., race, gender). Yet, fairness interventions often require exactly the sort of sensitive information that privacy mechanisms try to hide. In addition, privacy and fairness can inadvertently impact each other, bringing new privacy-fairness-utility trade-offs.
- Specifically, the interplay between privacy and fairness is both crucial and nuanced:

- Satisfying DP can unintentionally worsen fairness if certain subpopulations are more sensitive to noise [4].
- Enforcing fairness can unintentionally worsen privacy [5].

Therefore, the aim of this PhD thesis is to design, analyze, and implement differential privacy mechanisms that foster equitable ML outcomes while preserving model performance (utility). A key objective will then be to design a comprehensive open-source Python framework, offering ready-to-use building blocks for private and fair machine learning.

**Research Objective.** The goal of this PhD thesis is to advance the state of the art at the intersection of privacy and fairness by designing new representation learning techniques under LDP that enable fair and effective ML models. In particular, we aim to develop new locally private representations that preserve the utility of the data for downstream tasks while facilitating fair learning, even when sensitive attributes are obfuscated or partially unavailable.

The thesis will focus on:

- Designing new LDP-based data representations or embeddings optimized for fairness-aware ML.
- Theoretically analyzing the privacy-fairness-utility trade-offs of these