Dr Haiping Lu received the prestigious Amazon Research Award (ARA) on 15 January 2019. $61,000 USD in funding and an additional $10,000 USD in AWS (Amazon Web Services) Promotional Credits will support Haiping’s research student, Yan Ge, for one year working on the research project. As well as receiving funding recipients of the award are also invited to speak at Amazon offices worldwide about their work.
The research will look at developing tensor-based representation learning models and scalable algorithms that can preserve higher-order structures in networks.
I am really excited and grateful for receiving this prestigious award and the opportunities it brings. This recognition from Amazon demonstrates the relevance of our fundamental research to the real-world challenges. This award will deepen our ties with the Amazon Research teams and other industry partners. Their feedback will greatly help us assess the applicability of our research to real-world problems. It also provides invaluable exposure to our students via directly participating in the research or indirectly working on related topics covered in various modules. This again confirms Sheffield to be at the forefront of Machine Learning and Artificial Intelligence research. And I am really proud to be here and be able to contribute
Dr Haiping Lu
ARA winner
About the project
Learning Representations of Higher-Order Structures for Networks via Tensor Embedding
Learning representations of higher-order structures is key in analysing complex networks such as social networks and knowledge graphs to better understand interactions between their entities, e.g., for community detection, sentiment classification, and recommendation. Most existing methods analyse networks based on pairwise relations only, yet higher-order structures are important to the structure and function of complex networks. Recent works encode higher-order structures as a tensor (multidimensional array) for clustering, but tackle associated challenges with heuristics susceptible to information loss. We propose a tensor-based representation learning approach for networks that generalises recent network embedding models to higher orders. We will develop scalable algorithms for model parameter learning and explore clustering, link prediction, and node classification tasks. The outcomes can benefit network/graph mining applications and help customers find products more easily.