BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:GraphNeuralRAG: On the Opportunities and Challenges of GNNs for Gr
 aphRAG\, from Multi-Hop Question Answering to Perturbation Modelling - And
 rea Giuseppe Di Francesco\, Sapienza University of Rome\, ISTI-CNR
DTSTART:20260409T140000Z
DTEND:20260409T150000Z
UID:TALK245977@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:Retrieval-augmented generation (RAG) has become the standard a
 pproach for grounding generative models in external knowledge. When that k
 nowledge is structured as a graph\, GraphRAG methods have emerged to lever
 age topology to boost retrieval. However\, existing approaches predominant
 ly rely on either LLM-based pipelines\, which treat graph structure as tex
 t to summarise or traverse\, or use graph algorithms and neural scoring on
 ly as an intermediate step before falling back to document-based retrieval
 \, leaving much of the graph structure unexploited by the generative model
 . Graph Neural Networks (GNNs) offer a compelling middle ground: natively 
 designed for graph-structured data\, learnable end-to-end\, and capable of
  encoding complex relational patterns into compact representations. This t
 alk explores the opportunities and challenges of placing GNNs at the core 
 of GraphRAG pipelines\, presenting a GraphNeuralRAG framework across two d
 omains. In multi-hop question answering over knowledge graphs\, GNN-based 
 retrieval is shown to address key limitations of document-centric approach
 es\, including recall-precision tradeoffs\, token inefficiency\, and loss 
 of structural information. In perturbation modelling\, PT-RAG (ICLR 2026 W
 orkshop on Generative AI in Genomics) is presented as the first RAG framew
 ork for predicting single-cell responses to gene perturbations\, followed 
 by a discussion of how gene-gene networks provide a natural graph substrat
 e for extending this into a full GraphNeuralRAG pipeline. The focus throug
 hout is on both what GNNs uniquely enable and what remains hard\, offering
  an honest map of this emerging research direction.
LOCATION:Computer Laboratory\, William Gates Building\, Room LT1
END:VEVENT
END:VCALENDAR