Understanding Bottlenecks of State Space Models through the Lens of Recency and Over-smoothing

ICLR 2025 Conference Submission1195 Authors

Keywords: SSM, Locality, Recency, Over-smoothing

Abstract:

Structured State Space Models (SSMs) have emerged as alternatives to transformers, addressing the challenges of processing long sequences.
While SSMs are often regarded as effective in capturing long-term dependencies, we theoretically demonstrate that they suffer from a strong recency bias.
Our empirical findings reveal that this bias impairs the models' ability to recall distant information and introduces robustness issues.
We conducted scaling experiments and discovered that deeper structures in SSMs facilitate the learning of long contexts.
However, our theoretical analysis reveal that as SSMs increase in depth, they exhibit a tendency toward over-smoothing, resulting in token representations becoming increasingly indistinguishable.
This over-smoothing phenomenon ultimately constrains the scalability of SSMs to achieve improved performance.
Collectively, these findings highlight important limitations of SSMs and underscore the need for further research to address these challenges in long-range sequence modeling.

Primary Area: foundation or frontier models, including LLMs

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1195