The Diffusion-Attention Connection
Fuente:
arXiv
Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917399886823424 |
|---|---|
| author | Candanedo, Julio |
| author_facet | Candanedo, Julio |
| contents | Transformers, diffusion-maps, and magnetic Laplacians are usually treated as separate tools; we show they are all different regimes of a single Markov geometry built from pre-softmax query-scores. We define a QK "bidivergence" whose exponentiated and normalized forms yield attention, diffusion-maps, and magnetic diffusion. And use product of experts and Schrödinger-bridges to connect and organize them into equilibrium, nonequilibrium steady-state, and driven dynamics. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_09560 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | The Diffusion-Attention Connection Candanedo, Julio Machine Learning Transformers, diffusion-maps, and magnetic Laplacians are usually treated as separate tools; we show they are all different regimes of a single Markov geometry built from pre-softmax query-scores. We define a QK "bidivergence" whose exponentiated and normalized forms yield attention, diffusion-maps, and magnetic diffusion. And use product of experts and Schrödinger-bridges to connect and organize them into equilibrium, nonequilibrium steady-state, and driven dynamics. |
| title | The Diffusion-Attention Connection |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2604.09560 |