BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sheaf-Based Diffusion for Multimodal Graph Learning - Mar Gonzàle
 z i Català (Universitat Politècnica de Catalunya)
DTSTART:20250513T150000Z
DTEND:20250513T154500Z
UID:TALK232048@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:Multimodal Graph Learning (MGL) is an emerging area in machine
  learning that focuses on graphs whose nodes carry information from differ
 ent modalities\, such as text and image. A central challenge in MGL is int
 egrating these heterogeneous data types\, which are not directly comparabl
 e. Standard Graph Neural Networks (GNNs) struggle in multimodal contexts b
 ecause they assume homogeneity in node features and tend to merge modaliti
 es too early\, leading to the loss of valuable\, modality-specific informa
 tion. Existing solutions address this by processing each modality independ
 ently and fusing their predictions at the output level. However\, recent s
 tudies show that these late-fusion strategies underperform compared to gen
 eral-purpose GNNs. To address this limitation\, we introduce MMSheaf\, a f
 amily of sheaf-based neural network architectures that preserve modality s
 eparation before diffusion and introduce structured\, learnable mechanisms
  for cross-modal interaction during message passing. As a first contributi
 on\, we show that Sheaf Neural Networks (SNNs) outperform standard GNNs li
 ke GCN or GAT on multimodal graphs\, proving to be an appropriate tool for
  this context. Building on this insight\, our MMSheaf architecture further
  improves performance by explicitly modeling cross-modal interactions. We 
 evaluate MMSheaf on synthetic multimodal datasets where successful classif
 ication requires integrating modalities in a non-trivial way. Additional e
 xperiments on the real-world Ele-Fashion dataset showcase the model's effe
 ctiveness in practical multimodal settings. Overall\, our findings establi
 sh sheaf-based diffusion as a powerful and expressive framework for Multim
 odal Graph Learning. Future work will apply this approach to diverse domai
 ns such as biomedicine and recommender systems.\n\nMeet link: meet.google.
 com/wtt-wydt-hfk
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR