Anisotropy Is Inherent to Self-Attention in Transformers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Godey, Nathan, de la Clergerie, Éric, Sagot, Benoît
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910306788179968
author Godey, Nathan
de la Clergerie, Éric
Sagot, Benoît
author_facet Godey, Nathan
de la Clergerie, Éric
Sagot, Benoît
contents The representation degeneration problem is a phenomenon that is widely observed among self-supervised learning methods based on Transformers. In NLP, it takes the form of anisotropy, a singular property of hidden representations which makes them unexpectedly close to each other in terms of angular distance (cosine-similarity). Some recent works tend to show that anisotropy is a consequence of optimizing the cross-entropy loss on long-tailed distributions of tokens. We show in this paper that anisotropy can also be observed empirically in language models with specific objectives that should not suffer directly from the same consequences. We also show that the anisotropy problem extends to Transformers trained on other modalities. Our observations suggest that anisotropy is actually inherent to Transformers-based models.
format Preprint
id arxiv_https___arxiv_org_abs_2401_12143
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Anisotropy Is Inherent to Self-Attention in Transformers
Godey, Nathan
de la Clergerie, Éric
Sagot, Benoît
Computation and Language
The representation degeneration problem is a phenomenon that is widely observed among self-supervised learning methods based on Transformers. In NLP, it takes the form of anisotropy, a singular property of hidden representations which makes them unexpectedly close to each other in terms of angular distance (cosine-similarity). Some recent works tend to show that anisotropy is a consequence of optimizing the cross-entropy loss on long-tailed distributions of tokens. We show in this paper that anisotropy can also be observed empirically in language models with specific objectives that should not suffer directly from the same consequences. We also show that the anisotropy problem extends to Transformers trained on other modalities. Our observations suggest that anisotropy is actually inherent to Transformers-based models.
title Anisotropy Is Inherent to Self-Attention in Transformers
topic Computation and Language
url https://arxiv.org/abs/2401.12143