Is Anisotropy Inherent to Transformers?
This work addresses a fundamental issue in self-supervised learning for researchers and practitioners using Transformers across NLP and other domains, suggesting anisotropy might be a broader, inherent property rather than a specific optimization artifact.
The paper investigates whether anisotropy, a representation degeneration problem where hidden representations become unexpectedly close in angular distance, is inherent to Transformer-based models. It empirically shows anisotropy occurs in language models with objectives not directly linked to cross-entropy loss on long-tailed token distributions and extends to Transformers trained on other modalities.
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 tend to demonstrate that anisotropy might actually be inherent to Transformers-based models.