Analyzing Transformers in Embedding Space

Guy Dar, Mor Geva, Ankit Gupta, Jonathan Berant

פרסום מחקרי: פרק בספר / בדוח / בכנספרסום בספר כנסביקורת עמיתים

תקציר

Understanding Transformer-based models has attracted significant attention, as they lie at the heart of recent technological advances across machine learning. While most interpretability methods rely on running models over inputs, recent work has shown that an input-independent approach, where parameters are interpreted directly without a forward/backward pass is feasible for some Transformer parameters, and for two-layer attention networks. In this work, we present a conceptual framework where all parameters of a trained Transformer are interpreted by projecting them into the embedding space, that is, the space of vocabulary items they operate on. Focusing mostly on GPT-2 for this paper, we provide diverse evidence to support our argument. First, an empirical analysis showing that parameters of both pretrained and fine-tuned models can be interpreted in embedding space. Second, we present two applications of our framework: (a) aligning the parameters of different models that share a vocabulary, and (b) constructing a classifier without training by “translating” the parameters of a fine-tuned classifier to parameters of a different model that was only pretrained. Overall, our findings show that at least in part, we can abstract away model specifics and understand Transformers in the embedding space.

שפה מקוריתאנגלית
כותר פרסום המארחLong Papers
מוציא לאורAssociation for Computational Linguistics (ACL)
עמודים16124-16170
מספר עמודים47
מסת"ב (אלקטרוני)9781959429722
סטטוס פרסוםפורסם - 2023
אירוע61st Annual Meeting of the Association for Computational Linguistics, ACL 2023 - Toronto, קנדה
משך הזמן: 9 יולי 202314 יולי 2023

סדרות פרסומים

שםProceedings of the Annual Meeting of the Association for Computational Linguistics
כרך1

כנס

כנס61st Annual Meeting of the Association for Computational Linguistics, ACL 2023
מדינה/אזורקנדה
עירToronto
תקופה9/07/2314/07/23

ASJC Scopus subject areas

  • ???subjectarea.asjc.1700.1706???
  • ???subjectarea.asjc.3300.3310???
  • ???subjectarea.asjc.1200.1203???

טביעת אצבע

להלן מוצגים תחומי המחקר של הפרסום 'Analyzing Transformers in Embedding Space'. יחד הם יוצרים טביעת אצבע ייחודית.

פורמט ציטוט ביבליוגרפי