Enregistré dans:
| Auteurs principaux: | , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2512.23410 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915766065954816 |
|---|---|
| author | Kalyoncuoglu, Yusuf Miftachov, Ratmir |
| author_facet | Kalyoncuoglu, Yusuf Miftachov, Ratmir |
| contents | State-of-the-art models rely on massive widths despite exhibiting low Intrinsic Dimension (ID). We posit that this redundancy serves the non-convex optimization search rather than the final representation. We validate this hypothesis by decoupling the solution geometry via data-independent random projections, demonstrating that ResNet, ViT, and BERT representations can be compressed by up to 16x with negligible performance degradation of around 1%. Notably, these oblivious projections achieve parity with PCA and learned baselines, confirming the solution manifold is intrinsically robust. These findings establish the foundation for Subspace-Native Distillation: a paradigm where student models target this intrinsic manifold directly, bypassing the high-dimensional optimization bottleneck to realize the vision of "Train Big, Deploy Small" |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_23410 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | High-Dimensional Search, Low-Dimensional Solution: Decoupling Optimization from Representation Kalyoncuoglu, Yusuf Miftachov, Ratmir Machine Learning Artificial Intelligence State-of-the-art models rely on massive widths despite exhibiting low Intrinsic Dimension (ID). We posit that this redundancy serves the non-convex optimization search rather than the final representation. We validate this hypothesis by decoupling the solution geometry via data-independent random projections, demonstrating that ResNet, ViT, and BERT representations can be compressed by up to 16x with negligible performance degradation of around 1%. Notably, these oblivious projections achieve parity with PCA and learned baselines, confirming the solution manifold is intrinsically robust. These findings establish the foundation for Subspace-Native Distillation: a paradigm where student models target this intrinsic manifold directly, bypassing the high-dimensional optimization bottleneck to realize the vision of "Train Big, Deploy Small" |
| title | High-Dimensional Search, Low-Dimensional Solution: Decoupling Optimization from Representation |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2512.23410 |