An equality condition for the Dobrushin bound on attention rollout and how often it holds in trained transformers VBS62KXK7J
| URI | m2://res/VBS62KXK7J |
|---|---|
| Kind | preprint |
| Creators | Przemysław Rola |
| Date | 2026-10-04 |
| Subjects | arxiv cs.LG, arxiv math.PR |
| Scope | border |
Expressions
e1: version, preprint
| Date | 2026-10-04 |
|---|---|
| Ids | arXiv 2610.05558 |
Items in this resource
The mentions of the inventory that name this resource. Each name links to the card of its inventory row.
| Item | Kind | Where | Source |
|---|---|---|---|
| Exactness | statement | proposition 2 | arxiv |
| Factorwise tightness | statement | corollary 1 | arxiv |
| Residual bound; at $\alpha=\tfrac12$ | statement | lemma 1 | arxiv |
| The bound can fail completely | example | example 3 | arxiv |
Catalogue records
- arxiv: 2610.05558 (work)