서브 헤더

공지사항

공지사항

행사세미나 전문가 초청 세미나 개최 안내(Prof. Karl Stratos, Assistant professor @ Rutgers Uni…

페이지 정보

profile_image
작성자 관리자 댓글 0건 조회 1,311회 작성일 22.09.07

본문

전문가 초청 세미나 개최 안내드리오니 많은 관심과 참여 부탁드립니다. 감사합니다. 


Title: Retrieval-Based Approaches for Knowledge-Intensive Language Processing


Speaker: Prof. Karl Stratos, Assistant professor @ Rutgers University


Time : 2022 Sep 20th 10:00 ~ 11:00


Location: Online

https://us02web.zoom.us/j/86948464146  (비밀번호 0920)


After the talk, we will have an informal Q&A session with Prof. Karl Stratos. Please leave any questions about your research or career.


Abstract

Training a machine learning system to retrieve and read relevant parts of a knowledge base (KB), where a KB can be as general as a set of text blocks (e.g., passages in Wikipedia), is a relatively new paradigm (“open-domain”) in NLP that is dramatically changing how we approach language tasks. Compared to systems without a retrieval module, it has the following critical benefits. First, the model can handle unknown facts simply by consulting a (potentially new) KB that defines these facts at test time, yielding a natural zero-shot learning framework. Second, the model output is substantially more faithful in its information content because it conditions on precise statements, alleviating one of the most stubborn problems in modern NLP of hallucinating fake facts. Third, the model prediction is significantly more interpretable because we can examine what KB entries it accesses to make the prediction. For these reasons, there has been a surge of interest in framing knowledge-intensive language tasks as retrieve-and- read problems spanning QA, dialogue, fact checking, entity linking, and slot filling.

In this talk, I will present our recent projects on advancing retrieval-based approaches in NLP: (1) hard-negative noise contrastive estimation (NAACL 2021), (2) entity linking as question answering (ICLR 2022), and (3) multi-tasking retrieval (work in progress).



Bio:

Karl Stratos is an assistant professor in the Computer Science Department at Rutgers University and a Naver Scholar at Naver Corporation. His research centers on statistical approaches to unsupervised learning in natural language processing. He completed a PhD in computer science from Columbia University in 2016. After PhD, he was a senior research scientist at Bloomberg LP (2016-2017) and a research assistant professor at Toyota Technological Institute at Chicago (2017-2019).