Názov:Towards Explainable Neural Language Models
Vedúci:Mgr. Endre Hamerlik
Kµúčové slová:Explainable AI, xAI, Natural Language Processing, NLP, Transformer, probing
Abstrakt:Natural language processing has been rapidly advancing in recent years. The used neural models in this domain are more and more complex. Different techniques have been developed to understand their inner representations. Inspired by these techniques, we tried to extract some linguistic features from the inner representations of a chosen model, more precisely the BERT model. We conducted an ad hoc syntactic dependency analysis and used the so-called probing technique to predict parts of speech. We also present our results and discuss possible further work.

Súbory diplomovej práce:

Súbory prezentácie na obhajobe: