Projects

Neural networks are remarkably good at processing and drawing conclusions from raw data like images or text. In contrast, it is not yet clear how more processed, high level knowledge can be leveraged to augment traditional end-to-end networks.

HDMaps are a crucial component of modern trajectory forecasting techniques for autonomous driving. Their integration in neural networks is however not completely straightforward and requires further investigation.

Federated learning is an interesting paradigm for distributed learning that comes with its own set of new challenges. I am personally interested in the question of how updates from multiple clients should be aggregated to obtain the best training possible.

The MultiTrans project (funded by Agence National de la Recherche under grant reference ANR-21-CE23-0032) is a joint project between Insa Rouen, Valeo.ai and Inria Sophia Antipolis. The idea is to enable data, experience and knowledge to be transferable across the different systems (simulation, robotic models, and real-word cars), thus potentially accelerating the rate an embedded intelligent system can gradually learn to operate at each deployment stage. Existing autonomous vehicles are able to learn how to react and operate in known domains autonomously but research is needed to help these systems during the perception stage, allowing them to be operational and safer in a wider range of situations. MultiTrans proposes to address this issue by developing an intermediate environment that allows to deploy algorithms in a physical world model, by re-creating more realistic use cases that would contribute to a better and faster transfer of perception algorithms to and from a real autonomous vehicle test-bed and between multiple domains. I am currently working as postdoc on this project!

Multi-Input Multi-Output training leads to the emergence of smaller subnetworks within the base network. This has a number of benefits such a better utilization of network parameters, better training and strong regularization of learned features.

While Mixing Samples Data Augmentation (MSDA) offer a powerful tool for regularization, they offer very little control on the content mixed. I study how manipulating the contents mixed into the final augmented image can improve the training of networks.

Mixing Samples Data Augmentation (MSDA) help regularize models by mixing the contents of multiple images during training. I am interested in how these methods can help the models learn better and more general features by contrasting the contents of the original mixed inputs.