Missions

AI is well on the way to becoming indispensable in all engineering and service sectors. It is already making it possible to take advantage of large volumes of data available or provided by various entities (e.g., organizations, people, connected objects). However, making this data available raises issues of confidentiality and privacy, so federated learning approaches enable collaborative training of AI models, without data transmission. A central entity is in charge of coordinating the training of a global model, based on models received from client entities. These models are themselves derived from data that remains locally stored and does not need to be communicated.
Securing these approaches poses a number of problems relating to scaling and to attacks that malicious entities (e.g., clients) could carry out (e.g., by providing malicious models to guide the global model). The latter are essentially aimed at misleading or hijacking the global model (adversarial attacks). Membership inference attacks (MIA), for example, involve gathering information about the data used to produce the AI model. Reconstruction attacks, on the other hand, attempt to recover private data from aggregated information. While the former assume a dishonest or compromised central entity, the latter may simply analyze the parameter update shared from the local training (e.g. the local gradient or weight update vector), in order to reconstruct the private local training data.

Activities

The work being carried out at LAMIH (UMR 8201) is initially aimed at securing federated learning approaches in the face of these incident risks. It will then study the extent to which federated learning (or other forms) can contribute to securing infrastructures and services. The research will be carried out in collaboration with IRL ILLS (Montreal, Canada), where missions can be carried out.

Skills

• PhD degree in Computer Science or related field
• Experience in research around AI systems (machine learning) or cyber security
• Experience in programming languages (Python, C/C++) and scripting (Bash, Shell)
• Experience in Deep Learning frameworks (Pytorch/Tensorflow/MXNet)

Work Context

LAMIH UMR CNRS 8201 (Laboratoire d'Automatique, de Mécanique et d'Informatique Industrielles et Humaines) (Valenciennes) is a joint research unit of the Université Polytechnique Hauts de France (UPHF) and the Centre National de la Recherche Scientifique (CNRS), specialized in transportation and human mobility.
LAMIH is a multi-disciplinary laboratory and a recognized research player in the fields of transport and mobility: non-polluting vehicles, intelligent transport, driving aids, eco-driving, lighter structures, transport logistics, mobility for all and intelligent mobility.
It has strong expertise in all aspects of human interaction with technical systems.
The laboratory is organized into 4 departments:
• Automation,
• Computer Science,
• Mechanical Engineering,
• Human and Life Sciences (SHV).
This PhD project will be led within the Computer Science department.

The position is located in a sector under the protection of scientific and technical potential (PPST), and therefore requires, in accordance with the regulations, that your arrival is authorized by the competent authority of the MESR.

Constraints and risks

The position is located in a sector falling under the protection of scientific and technical potential (PPST), and therefore requires, in accordance with regulations, that your arrival be authorized by the competent authority of the MESR.
Taking into account the specificities of the unit, closures are imposed by the UPHF: 4 weeks for the summer closure, 2 weeks at the end of the year, 1 week during the winter holidays, 1 week during the spring holidays. The recruited agent will be required to take leave during these periods.