BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260924T232456EDT-1388BANNGz@132.216.98.100
DTSTAMP:20260925T032456Z
DESCRIPTION:Abstract\n\nThis thesis comprises three interconnected projects
  addressing challenges in multi-label classification and temporal graph le
 arning. In the first project\, we tackle the challenge of modelling label 
 dependencies in multi-label classification. We introduce a graph-based dep
 endency module that is capable of modeling multiple types of relations. Th
 e module can be incorporated in embedding-based multi-label classification
  methods\, leading to Relation Guided Message Passing (RGMP)\, a novel mul
 ti-label classification approach. We demonstrate via experiments that the 
 proposed method achieves superior or comparable performance to state-of-th
 e-art methods across all studied datasets\, without imposing substantial a
 dditional model complexity or computational overhead. This emphasizes the 
 importance of capturing diverse label dependencies.\n\nSecondly\, we addre
 sses multi-label text classification in annotation-free and scarce-annotat
 ion settings. Our method leverages pre-trained language models for natural
  language inference\, constructs a signed label dependency graph\, and uti
 lizes message passing along this graph to generate effective label predict
 ions. In the weak supervision setting\, where we have access to only a ver
 y small set of labelled data\, our approach achieves significant performan
 ce improvement compared to existing techniques.\n\nFinally\, we introduce 
 the task of recent link classification\, which is important in industrial 
 settings but has received little attention from the research community. In
  this task the goal is to predict the label of a recently observed edge be
 tween nodes. This problem arises when we observe an interaction between tw
 o entities (e.g.\, a potentially fraudulent transaction in a financial net
 work)\, but will not have access to the label of the interaction until muc
 h later. We outline how this task can act as a benchmark task for evaluati
 ng Temporal Graph Learning (TGL) methods. We formalize the task\, propose 
 benchmark datasets\, and evaluate state-of-the-art methods using robust me
 trics. We demonstrate how modifications in message aggregation\, readout l
 ayer\, and time encoding strategies can yield substantial performance impr
 ovement. Additionally\, we present a novel learning architecture (Graph Pr
 ofiler)\, capable of encoding previous events’ class information\, achievi
 ng enhanced performance on most cases of interest.\n
DTSTART:20240314T140000Z
DTEND:20240314T160000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Muberra Ozmen – Graph-based strategies for classific
 ation with diverse label information
URL:https://www.mcgill.ca/ece/channels/event/phd-defence-muberra-ozmen-grap
 h-based-strategies-classification-diverse-label-information-355998
END:VEVENT
END:VCALENDAR