Internationally recognized generalist engineering school of the IMT (Institut Mines-Télécom), leading French engineering school (Technological University), IMT Atlantique aims to support transitions, train responsible engineers, and use scientific excellence to serve teaching, research and innovation.

This position is attached to the Data Science Department, part of whose research activity focuses on data security (images, databases, 3D surfaces, etc.) and processing (e.g. federated AI), with issues such as digital forensics, the fight against information leaks, federated learning attacks and defenses, and so on. These same objectives are also being addressed in the context of data storage on DNA molecules.

This position is part of CARBURE, a DGA Rapid project in partnership with eXo maKina, and aims to make decision-making more explicable in the fight against fake/edited image using deep neural networks.

In the case of a fake image, an expert must carry out an appropriate analysis and justify his or her decision as to whether or not an image has been edited. To do this, he or she needs to conduct a detailed image analysis using measurements from several filters. The Tungstène software, developed by eXo maKina, provides access to "filters" which offer different views on the nature and composition of images, without however providing any decision, even partial, on the generated or edited nature of the image: this is the objective of CARBURE.

MISSIONS

Under the functional responsibility of Pr. Gouenou Coatrieux and in collaboration with researchers from eXo maKina, the main tasks of the position are:

• Complete a bibliography
• Propose a justified decision aid support
• Develop a dedicated application
• Promote research results

ACTIVITIES:

Consolidate a bibliographical and theoretical work

• Carry out bibliographical and theoretical work on the explicability of decision-making based on deep networks.

Improve a justified decision support algorithm (thematic, not network-specific) to

• Make decision-making faster and more consistent
• Take as input of a deep algorithm the Tungsten's various filters, rather than the initial image, so as to make the decision-making process coherent with the analysis that an expert could provide.

Develop a dedicated application

• Develop a dedicated application based on these networks. A database of edited images will be provided (specifying the types of images, photographic or generated, and the associated types of alteration), enabling appropriate training.

Promote research results

• Provide reports and project deliverables in collaboration with line managers.
• Writing of scientific publications resulting from research carried out in collaboration researchers involved in the project.
• Participate in the dissemination and valorization of results.
• Participate in meetings related to the mission.

Minimum education and/or experience required:

• PhD obtained less than 3 years before date of hire in computer sciences

Essential skills, knowledge and experience:

• Skills/knowledge in machine learning, deep learning,
• Skills/knowledge in neural network explainability would be appreciated.
• Development under Linux or Windows with python / pyTorch mainly.
• Experimentation fundamentals
• Writing skills in French and (especially) in English absolutely essential.

Abilities and skills:

• Autonomy and initiative
• Teamwork skills
• Professional written and spoken expression

APPLICATION DOCUMENT

• CV
• cover letter
• recommendation letter

Information request

• Gouenou Coatrieux - Professor - gouenou.coatrieux@imt-atlantique.fr
• Grégoire Mercier -R&D Director, eXO maKina - gregoire.mercier@exomakina.fr
• For administrative/HR aspects: Aurore Forny - aurore.forny@imt-atlantique.fr

Deadline for application : 2024, december 20th

Start of the contract: 2025, February

Interviews : As soon as possible