Length independent generalization bounds for deep SSM architectures via Rademacher contraction and stability constraints
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913856257785856 |
|---|---|
| author | Rácz, Dániel Petreczky, Mihály Daróczy, Bálint |
| author_facet | Rácz, Dániel Petreczky, Mihály Daróczy, Bálint |
| contents | Many state-of-the-art models trained on long-range sequences, for example S4, S5 or LRU, are made of sequential blocks combining State-Space Models (SSMs) with neural networks. In this paper we provide a PAC bound that holds for these kind of architectures with \emph{stable} SSM blocks and does not depend on the length of the input sequence. Imposing stability of the SSM blocks is a standard practice in the literature, and it is known to help performance. Our results provide a theoretical justification for the use of stable SSM blocks as the proposed PAC bound decreases as the degree of stability of the SSM blocks increases. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_20278 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Length independent generalization bounds for deep SSM architectures via Rademacher contraction and stability constraints Rácz, Dániel Petreczky, Mihály Daróczy, Bálint Machine Learning Artificial Intelligence 68 I.2.6 Many state-of-the-art models trained on long-range sequences, for example S4, S5 or LRU, are made of sequential blocks combining State-Space Models (SSMs) with neural networks. In this paper we provide a PAC bound that holds for these kind of architectures with \emph{stable} SSM blocks and does not depend on the length of the input sequence. Imposing stability of the SSM blocks is a standard practice in the literature, and it is known to help performance. Our results provide a theoretical justification for the use of stable SSM blocks as the proposed PAC bound decreases as the degree of stability of the SSM blocks increases. |
| title | Length independent generalization bounds for deep SSM architectures via Rademacher contraction and stability constraints |
| topic | Machine Learning Artificial Intelligence 68 I.2.6 |
| url | https://arxiv.org/abs/2405.20278 |