Stage - Ontology Aided Neuro-Symbolic Reasoning With Llms H/F CEA

Rejoignez-nous en Stage !

The latest advances on AI and LLMs (and their variations) mean a next stage on technological
advancements. These technologies in turn, lead to new services, products and interactions, which are
currently shaping and impacting businesses, industries and people on daily basis, and at an
increasing pace. Yet, in part due to the fast advancement, in part due to the extraordinary tasks
accomplished by these technologies, the real capabilities and limits of LLMs along with their
integration with already existing approaches is still being unveiled.
This position is aimed at exploring and experimenting on hybrid (Neuro-Symbolic) technologies where
the interaction of LLMs and Symbolic AI is studied. Specifically, the work will be framed by the use of
Ontologies (Konwledge Representation artifacts) as domain specific knowledge, that guides, restricts
and conditions the interaction and responses of the LLMs/VLMs.
The integration of symbolic AI contibutes with traceability, interpretability, proveability and well
understood semantics, which we intend to propagate to generative AI solutions.
Internship Objectives
- Study the SOA and approaches on LLMs/VLMs + reasoning, with emphasis on Ontologies and
other formal Knowledge Representation approaches.
- Research on the ways the interaction can be exploited and provide added value (e.g. conformity).
- Propose, produce and integrate into experiments that demonstrate the principles.
- Redact, propose, refine theoretical background supporting and explaining the claims.
The position is hybrid :
- Reading, research and writing