Le prochain séminaire LISIC sera donné par Zied Bouraoui du laboratoire CRIL de l’Université d’Artois le jeudi 23 novembre à 14h en B014.
Learning Semantic Concept Embedding from Langage Models
Learning static vectors that capture the meaning of concepts remains a fundamental challenge in applications where word meaning has to be modelled in the absence of (sentence) context. For instance, static word vectors are needed for zero-shot image classification, and zero-shot entity typing, for ontology alignment and completion, taxonomy learning, or for representing query terms in information retrieval systems. In this talk, I will discuss some strategies that can be pursued to learn effective concept representations from language models. I will also provide an overview of how downstream applications could benefit from such embeddings.