In recent years, machine learning approaches, and in particular deep neural networks, have yielded significant improvements on several natural language processing and computer vision tasks; however, such breakthroughs have not yet been observed in the area of information retrieval (IR). Besides the complexity of the IR tasks, such as understanding the user's information needs, a main reason is the lack of large-scale training data for many IR tasks. This necessitates studying how to design and train machine learning algorithms where there is no large-scale data in hand. In this talk, I will introduce training neural networks for IR tasks with weak supervision, where labels are obtained automatically without human annotators or any external resources (e.g., click data). I will also present the application of such learning strategy in relevance ranking, word embedding, query performance prediction, and efficient learning to rank.

Hamed Zamani is a third-year PhD candidate (with distinction) in the Center for Intelligent Information Retrieval at the University of Massachusetts Amherst, working with W. Bruce Croft. His research interests include various aspects of core information retrieval, such as query representation, document representation, and ranking. His research mostly focuses on unsupervised or weakly supervised approaches. Hamed is an active member in the IR community. He is an ACM SIGIR student liaison, representing north and south Americas, and an organizer of the ACM RecSys Challenge workshop at the RecSys '18 conference. He has also served as a PC member at major IR (and related) venues, such as SIGIR, WSDM, WWW, RecSys, and CIKM.