Description: Real-world graphs often exist with multiple views, where each view describes a distinct type of interaction among a shared set of vertices. For example, in social networks, interactions between individuals include friendships, professional relationships, and various other types of relationships. Moreover, multiplex graphs are also prevalent in biology, such as protein-protein interactions gathered from various sources. In contrast to traditional single-view graphs, multiple views retain distinct semantics that complement one another. Hence, analyzing these multiple views collectively can offer a more holistic comprehension of the underlying graph, potentially unveiling patterns and relationships that might remain concealed when only examining a single view. In this project, we will investigate the limitations of existing techniques used for learning from multi-view graphs and create an efficient graph neural network framework for modelling these multi-view graphs, with the aim of enhancing downstream tasks such as gene function prediction.
Prerequisites: This project requires students to have a solid background in machine learning, algorithms, and excellent programming skills Python and PyTorch. ANU students are expected to have finished COMP4670 or have an equivalent experience.
Background literature:
[1] HDMI: High-order Deep Multiplex Infomax, Jing et al., The Web Conference (WWW), 2021.
[2] Multiplex Bipartite Network Embedding using Dual Hypergraph Convolutional Networks, Xue H, Yang L, et al., The Web Conference(WWW), 2021.
[3] Role-based Multiplex Network Embedding, Hegui Zhang and Gang Kou, The 39th International Conference on Machine Learning (ICML) 2022.
[4] Gemini: Memory-Efficient Integration of Hundreds of Gene Networks with High-Order Pooling, Woicik et al., Bioinformatics 2023.