DSC internal talk in September

Prof. Taro Watanabe gave a lecture in September.

The details are as follows.

====================

Prof. Taro Watanabe (Natural Language Processing Laboratory)

TITLE:
Language Model Knowledge Probing

ABSTRACT：

Language models have been employed in various knowledge-intensive tasks, like question-answering, and it is critical to investigate the factual knowledge encoded in language models. In this study, we propose an efficient knowledge probing method, DEcoder Embedding-based Relational (DEER), in which both queries and answers are encoded by language model internal representations, allowing efficient inference and robust probing.