Postdoctoral Researcher in Secure, Confidential, and Explainable Decentralized Learning - 12-month Fixed-Term Contract

About Télécom SudParis:

Télécom SudParis is a prestigious public engineering school recognized at the highest level in digital science and technology. The quality of its training is based on the scientific excellence of its faculty and a pedagogy emphasizing team projects, disruptive innovation, and entrepreneurship. Télécom SudParis has 1,000 students, including 700 engineering students and approximately 150 doctoral students. Télécom SudParis is part of the Institut Mines-Télécom, the leading group of engineering schools in France, and shares its campus with Institut Mines-Télécom Business School. Télécom SudParis is a co-founder of the Institut Polytechnique de Paris (IP Paris), a world-class science and technology institute with École Polytechnique, ENSTA Paris, ENSAE Paris, ENPC, and Télécom Paris. Video presentation of Télécom SudParis

About Institut Mines-Télécom:

The Institut Mines-Télécom (IMT) is a public institution dedicated to higher education and research for innovation in the fields of engineering and digital technology. Constantly attuned to the economic world, IMT combines strong academic and scientific legitimacy, proximity to businesses, and a unique positioning on the major transformations of the 21st century: digital, energy, industrial, and educational. Its activities are deployed within the Mines and Télécom engineering schools under the supervision of the minister in charge of Industry and Electronic Communications, two subsidiaries, and associated or affiliated partners. IMT is a founding member of the Alliance Industrie du Futur. It is dual-certified Carnot for the quality of its collaborative research. Video presentation of Institut Mines-Télécom

Missions

• Analyze the security, privacy, and ethical issues related to the deployment of artificial intelligence (AI) solutions in critical and diverse domains, particularly healthcare, finance, and transportation.

• Study privacy-preserving machine learning (PPML) methods and their capacity to meet regulatory and ethical requirements, especially those stemming from GDPR and the AI Act.

• Design and evaluate privacy protection approaches in decentralized learning environments, particularly within the framework of federated learning.

• Identify and characterize security vulnerabilities specific to collaborative and federated learning, including data poisoning attacks, backdoor injection, and attacks targeting data and user privacy.

• Study inference mechanisms from model updates, including membership inference and property inference attacks, to assess the risks of sensitive information disclosure.

• Evaluate the robustness and confidentiality of AI models in contexts where data, nodes, and learning processes are distributed and difficult to audit.

• Study explainable artificial intelligence (XAI) methods to improve the transparency, interpretability, and understanding of decisions made by AI models.

• Analyze the interactions and trade-offs between explainability, privacy, and robustness in decentralized learning systems.

• Develop and experiment with methods for reconciling transparency and privacy protection, taking into account the specific constraints of distributed environments.

• Implement experimental protocols and evaluation criteria to measure the levels of privacy, robustness, and explainability of the studied approaches.

• Contribute to the valorization and dissemination of research results, particularly through scientific publications, presentations at conferences, and participation in exchanges with project partners.

• Maintain scientific and technological watch on advancements in trustworthy AI.

Activities

• Design and develop decentralized learning mechanisms offering strong privacy guarantees, resistance to malicious behavior, and relevant explainability capabilities.

• Study the privacy of explainability mechanisms, particularly the risks of information leakage that may be induced by explainability methods applied to federated learning models.

• Develop new attack and evaluation strategies to characterize the information likely to be disclosed by explainability mechanisms.

• Design adapted protection mechanisms to limit identified information leaks while preserving the quality and relevance of the produced explanations.

• Develop collaborative learning approaches that reconcile privacy and explainability, taking into account the specific constraints of federated and decentralized environments.

• Study the use of Differential Privacy (DP) to protect aggregated models against inference attacks and analyze its impact on the quality and fidelity of explanations.

• Explore cryptographic primitives suitable for federated learning, including partially or fully homomorphic encryption and Secure Multi-Party Computation (MPC) techniques.

• Combine cryptographic mechanisms and differential privacy to explore trade-offs between privacy level, explanation quality, computational cost, and potential information leaks.

• Study the impact of privacy protection mechanisms on explainability methods, particularly attribution-based methods, and identify strategies to preserve their fidelity.

• Analyze the vulnerabilities of federated learning systems against Byzantine attacks and poisoning attacks, especially when adversarial strategies exploit metrics or information from explainability.

• Study realistic attack scenarios considering the heterogeneity and imbalance of data held by different clients, label poisoning, and the presence of clients with protected attributes.

• Develop robust and privacy-preserving aggregation algorithms capable of maintaining model performance in environments with a significant proportion of malicious participants.

• Design adaptive defense mechanisms leveraging explainability, particularly for identifying abnormal behavior, suspicious updates, or potentially malicious participants.

• Jointly analyze the trade-offs between privacy, robustness, explainability, and model utility, also integrating constraints related to computational efficiency and communication.

• Define evaluation protocols and metrics to compare proposed approaches according to different levels of privacy, robustness, explanation fidelity, and computational cost.

• Experiment and evaluate the developed methods on realistic decentralized learning scenarios, particularly in the healthcare domain and other contexts involving sensitive data.

• Contribute to the experimental validation and valorization of results, including the implementation of prototypes, analysis of results, drafting of scientific publications, and presentation of work at scientific events.

Education

• Doctorate or PhD in Computer Science obtained less than 3 years ago

Essential skills, knowledge, and experience

• Experience in AI or machine learning (ML) applied to cybersecurity

• Skills in Federated Learning

• Advanced mathematics skills

• English spoken and written

Desirable skills, knowledge, and experience

• Skills in homomorphic cryptography, particularly CKKS

• Skills in differential privacy

• Skills in Explainable AI (XAI)

• Skills in Privacy-Enhancing Technologies (PET)

Abilities and aptitudes

• Rigor, project management methodology

• Interpersonal skills with stakeholders at all levels, listening and cooperation skills

• Ability to work in a team

• Strict adherence to confidentiality obligations regarding the data used

Salary indications:

• Indicative salary range (excluding variable annual bonus): 35,400 - 38,000 euros gross per year, depending on profile and experience (reduced social charges in the public sector).

Additional information and application

• Application deadline: October 31, 2026

• Contract type: 12-month fixed-term contract

• Job category and position (internal use): II -P, Postdoctoral Researcher A (civil service)

• Position open to jobs at the immediately lower level (internal use)

• Location: Evry-Courcouronnes (91)

• Positions offered for recruitment are open to all, with, upon request, accommodations for candidates with disabilities.

• Employment open to civil service incumbents and/or contract employees.

• Working conditions: 44 days of leave, remote work possible, restaurant and cafeteria on site, accessibility by public transport (with employer contribution) or near major roads, staff association and sports association on campus.

Find more English Speaking Jobs in France on Arbeitnow