An encoder-centric view of retrieval

Day - Time: 23 April 2024, h.12:00
Place: Area della Ricerca CNR di Pisa - Room: Aula Faedo - C-29
  • Andrew Yates (IRLab, University of Amsterdam, The Netherlands)

Franco Maria Nardini

In this talk, I will describe my encoder-centric view of neural methods for retrieval and how different types of approaches compare under this framework. In some sense, traditional methods like BM25 are simply handcrafted encoders; in another, DSI (differentiable search index) is an approach to produce dense representations without any encoder. Motivated by the desire to gain more control over what is encoded in a query or document representation, I will describe how learned sparse representations can be adapted to a variety of settings. Throughout the talk, focusing on the architecture of the encoder will highlight similarities between methods and the importance of describing one's experimental pipeline in detail.

Andrew Yates is an Assistant Professor at the University of Amsterdam, where his research focuses on developing content-based neural ranking methods and leveraging them to improve search and downstream tasks. He has co-authored a variety of papers on neural ranking methods as well as a book on transformer-based neural methods: "Pretrained Transformers for Text Ranking: BERT and Beyond". Previously, Andrew was a post-doctoral researcher and senior researcher at the Max Planck Institute for Informatics. Andrew received his Ph.D. in Computer Science from Georgetown University, where he worked on information retrieval and extraction in the medical domain.

Meeting ID: 858 1373 8981 - Passcode: 662332