Explainable AI Team Seminar(Talk by Étienne Simon, University of Oslo).
RIKEN AIP online seminar: Étienne Simon (University of Oslo) on evaluating the compositional generalization of language models.
- When
- Wed, October 14, 2026 · 17:30–18:30 JST
- Where
- Online
- Organizer
- RIKEN Center for Advanced Intelligence Project
- Language
- EN
- Source
- Doorkeeper
Summary
An online seminar hosted by the Explainable AI Team at RIKEN AIP, featuring a talk by Étienne Simon of the University of Oslo titled "Evaluating Compositional Generalization of Language Models." It runs on Wednesday, October 14, 2026 from 17:30 to 18:30 JST.
Compositionality holds that the meaning of a complex expression comes from its parts and how they combine, but turning this into practical tests of language models is hard. The talk brings together three approaches: whether knowledge-graph-grounded models generalize to longer sequences and novel combinations (finding limits in both), systematic generalization across datasets linked to the entropy of component distributions in training data, and epsilon-compositionality, a continuous measure applied to text embeddings of idiomatic and non-idiomatic expressions.
About the community
A national AI research center in Japan that regularly opens its research team seminars to the public online. Invited researchers from universities around the world present recent work on machine learning theory, explainability, and language models, typically as one-hour English talks.
#explainable-ai#language-models#compositionality#nlp#research-seminar#riken-aip