BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Continual Learning on Dynamic Graphs and Its Applications
DTSTART:20220610T090000
DTEND:20220610T110000
DTSTAMP:20260504T083033Z
UID:883b1a748a77b1d85dc7d60c2180ee8a02f2396d74c19c3baaf0de71
CATEGORIES:Conferences - Seminars
DESCRIPTION:Shaobo Cui\nEDIC candidacy exam\nExam president: Prof. Matthia
 s Grossglauser\nThesis advisor: Prof. Boi Faltings\nThesis co-advisor: Pro
 f. Antoine Bosselut\nCo-examiner: Prof. Robert West\n\nAbstract\nCurrent g
 raph neural networks(GNNs) lack scalability when the graph evolves over ti
 me(addition/deletion of edges/nodes\, changes of edge weights\, etc.). Tra
 ining the whole graph from scratch every time step is time-consuming and n
 ot practical in real-world scenarios. Inspired by the fact that most alter
 nations influence only certain local parts of the whole graph while the ma
 jor parts remain the same\, we design an incremental learning approach for
  evolving graphs based on subgraphs. Specifically\, we relearn the influen
 ced subgraphs while retaining the ability on the uninfluenced part of the 
 whole graph.\nWe conduct experiments on a citation network dataset and pro
 ve the feasibility of our proposed method. Furthermore\, we design a more 
 precise continual learning approach that takes the evolving degree of subg
 raph into account.\n\nBackground papers\n\n	DyRep: Learning Representation
 s over Dynamic Graphs (https://par.nsf.gov/servlets/purl/10099025)\n	DyKg
 Chat: Benchmarking Dialogue Generation Grounding on Dynamic Knowledge Grap
 hs (https://arxiv.org/abs/1910.00610)\n	Graph Meta Learning via Local Sub
 graphs (https://arxiv.org/abs/2006.07889)\n
LOCATION:INR 212 https://plan.epfl.ch/?room==INR%20212
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR