From sensors to knowledge: Communicate and decideProjects
The team's projects are hosted at CERV, a technological platform from ENIB.
Current projects
HICAAR : Hippocampus-Inspired Computational model for general Artificial Abstract Reasoning (Inria Quadrant Programme – “High-Risk Research”)
The project aims to leverage recent neuroscience findings and deep learning architectures to address the challenge of abstract reasoning—the capacity to generalize skills by extracting general laws and transferring them across different tasks—in artificial intelligence.
This involves encoding abstract actions and perceptions and developing algorithms to learn resolutions at an abstract level. The methodology relies on twophases, with the start of the second phase conditioned by the success of the first. The first phase involves a precise and simple task that includes transferring reasoning from one example to another. In case of success with this task, we will address the core problem with complex examples provided by the Artificial General Intelligence community.
contact: pierre.deloor@enib.fr
Development of Evaluation Frameworks for AI in Brittonic Languages (with the support of the Brittany Region and the Welsh Government)
This network brings together Bangor University (Wales), the team of the ANR YAR project led by the CNRS IKER laboratory, and researchers from Lab-STICC. With the aim of advancing AI sovereignty for Brittonic languages, we seek to strengthen an ecosystem for the co-construction of foundational building blocks for the automatic processing of these languages.
This project focuses on evaluating the performance of AI models in our languages. We will develop methodologies to create very high-quality evaluation datasets in Breton and Welsh, using a human-centred evaluation approach. Evaluations will be carried out across different models, and their respective results will be analysed and presented in a clear, comparative, and accessible graphical form.
Welsh, Breton, and international open-source tools will be used and improved in order to deliver the project’s outputs.
contact: anne-gwenn.bosser@enib.fr