Networks, Machine Learning and Crime Analysis
Speaker: Kubilay Atasu
Time: 10:30 - 11:15
Location: Building 28: Turing room 0.E420
Zoom: link
Abstract: Graph machine learning (ML) is an emerging field of Artificial Intelligence motivated by the ubiquity of graph-structured data in real-life applications. Graph neural networks (GNNs) serve as a key technology in this area and can be leveraged in various fields including the financial industry. In this talk, I will cover applications of graph machine learning to financial crime analysis tasks such as anti-money laundering and phishing fraud detection.
Advances in Multigraph Neural Networks
Speaker: Çağrı Bilgi
Time: 11:15 - 11:35
Location: Building 28: Turing room 0.E420
Zoom: link
Abstract: Many real-world scenarios involve multigraphs, which can feature several parallel edges between the same pair of nodes. Existing multigraph neural network architectures either preprocess the multigraph by collapsing it into a simple graph before applying message passing or introduce auxiliary edge features that compromise the permutation equivariance property. We introduce MEGA-GNN, which overcomes these limitations by employing a two-stage aggregation process in the message passing layers: first, parallel edges between the same two nodes are aggregated, and then messages from distinct neighbors are combined. Our experiments show that MEGA-GNN significantly outperforms other multigraph neural network solutions across various real-world applications, such as anti-money laundering and phishing detection.
Higher-order temporal network prediction and interpretation
Speaker: Bart Peters
Time: 11:35 - 11:55
Location: Building 28: Turing room 0.E420
Zoom: link
Abstract: A social interaction (so-called higher-order event/interaction) can be regarded as the activation of a hyperlink among the corresponding individuals. Social interactions can be, thus, represented as higher-order temporal networks that record the higher-order events occurring at each time step over time. The prediction of higher-order interactions is usually overlooked in traditional temporal network prediction methods, where a higher-order interaction is regarded as a set of pairwise interactions. The prediction of future higher-order interactions is crucial to forecast and mitigate the spread of information, epidemics and opinion on higher-order social contact networks. In our work, we propose novel memory-based models for higher-order temporal network prediction. By using these models, we aim to predict the higher-order temporal network one time step ahead, based on the network observed in the past. Importantly, we also intend to understand what network properties and which types of previous interactions enable the prediction. The design and performance analysis of these models is supported by our analysis of the memory property of networks, e.g., similarity of the network and activity of a hyperlink over time, respectively. Our models assume that a target hyperlink’s future activity (active or not) depends on the past activity of the target link and of all or selected types of hyperlinks that overlap with the target. We then compare the performance of our models with three baseline models, which are an activity driven model, a probabilistic group-change model and a pairwise temporal network prediction method. In eight real-world networks, we find that both our models consistently outperform the baselines. Moreover, the refined model, which only uses a subset of all types of overlapping hyperlinks, tends to perform the best. Our models also reveal how past interactions of the target hyperlink and different types of hyperlinks that overlap with the target contribute to the prediction of the target’s future activity.