always curious about the world around me
During my PhD study, my research focus was initially on designing and implementing novel machine learning and deep learning algorithms for recommender systems that can handle massive datasets. Recently I have been more interested in natural language processing research, especially its application within financial markets. Moreover, I am quite interested in the interaction of computer vision and NLP, which I think is quite promising. My research works have been published at top machine learning conferences KDD, ICML, NeurIPS, AISTATS and CVPR. I also review papers regualrly for ICML, NeurIPS to contribute back to the community. I was jointly advised by Professor Cho-Jui Hsieh and Professor James Sharpnack.
More specifically, I have been working on Collaborative Ranking problem recently. Traditionally, people use matrix factorization approach for recommender systems. However, matrix factorization approach isn't minimizing the ranking loss directly. Therefore, recommendation accuracy using matrix factorization is not as good as using Collaborative Ranking Algorithms. The reason why Collaborative Ranking Algorithms have not been very popular yet is that the existing ones are very very very slow! Our proposed algorithms Primal-CR, Primal-CR++ are the first Collaborative Ranking Algorithms which are able to scale up to full Netflix dataset without sub-sampling on a single machine. If you are interested and want to know more, you can watch the youtube video, download the Julia code or C++ codes from Github, or read the KDD'17 paper. One can also refer to the slides and poster prepared for the talk I gave in Canada. (I accidentally found that KDD uploaded the video of my talk online already: here.) And I presented the work again at Amazon's Annual Graduate Research Symposium in Seattle, Washington during Oct 17-21, 2017.
For part of the year 2018, I have been working on a novel alternative listwise approach to Collaborative Ranking to supplement the KDD'17 pairwise approach. The new algorithm we proposed is called SQL-Rank, which stands for Stochastically Queuing Listwise Ranking Algorithm and has just been accepted to ICML'18 for oral presentation. I gave an oral presentation at Stockholm, Sweden with the slides here and poster here. One can read the pulished ICML'18 paper or play with the julia codes.
Advances in Collaborative Filtering and Ranking. Wu, Liwei. "Advances in Collaborative Filtering and Ranking." arXiv preprint arXiv:2002.12312 (2020).
SSE-PT: Sequential Recommendation Via Personalized Transformer. Wu, Liwei, et al. "Temporal Collaborative Ranking Via Personalized Transformer." arXiv preprint arXiv:1908.05435 (2019).
Multimodal Categorization of Crisis Events in Social Media. Mahdi*, Liwei*, et al.. "Multimodal Categorization of Crisis Events in Social Media." to appear in CVPR 2020. (* means equal contribution with ordering decided by Python)
Graph DNA: Deep Neighborhood Aware Graph Encoding for Collaborative Filtering. Wu, Liwei, et al. "Graph DNA: Deep Neighborhood Aware Graph Encoding for Collaborative Filtering." arXiv preprint arXiv:1905.12217 (2019).
Stochastic Shared Embeddings: Data-driven Regularization of Embedding Layers. Wu, Liwei, et al. "Stochastic Shared Embeddings: Data-driven Regularization of Embedding Layers." Advances in Neural Information Processing Systems. 2019.
SQL-Rank: A Listwise Approach to Collaborative Ranking. Wu, Liwei, Cho-Jui Hsieh, and James Sharpnack. "SQL-Rank: A Listwise Approach to Collaborative Ranking." Proceedings of Machine Learning Research (35th International Conference on Machine Learning). Vol. 80. 2018.
Large-scale Collaborative Ranking in Near-Linear Time. Wu, Liwei, Cho-Jui Hsieh, and James Sharpnack. Large-scale Collaborative Ranking in Near-Linear Time. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2017.
Template design by Andreas Viklund