Dextr: Zero-Shot Neural Architecture Search with Singular Value Decomposition and Extrinsic Curvature
Rohan Asthana [email protected] Friedrich-Alexander-Universität Erlangen-Nürnberg Erlangen, Germany Joschua Conrad [email protected] Universität Ulm Ulm, Germany Maurits Ortmanns [email protected] Universität Ulm Ulm, Germany Vasileios Belagiannis [email protected] Fr · 2025
Read Dextr: Zero-Shot Neural Architecture Search with Singular Value Decomposition and Extrinsic Curvature by Rohan Asthana [email protected] Friedrich-Alexander-Universität Erlangen-Nürnberg Erlangen, Germany Joschua Conrad [email protected] Universität Ulm Ulm, Germany Maurits Ortmanns [email protected] Universität Ulm Ulm, Germany Vasileios Belagiannis [email protected] Fr (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-12977.txt.
Description
Rohan Asthana, Joschua Conrad, Maurits Ortmanns, and Vasileios Belagiannis present Dextr, a zero shot way to pick neural network shapes with no training runs. Such searches rank shapes by cheap proxy scores read at the start. Their work came out in twenty twenty-five.
Most proxies need labeled data, which real settings often lack, and they stress either learning speed or expressive power alone. The team first shows how aligned channels affect both speed of learning and how well a network generalizes. Their proxy blends all three traits, built from a matrix split of layer features and the curve of the network output. It needs no labels, yet forecasts test scores well.
Catalog record
| CREATOR | Rohan Asthana [email protected] Friedrich-Alexander-Universität Erlangen-Nürnberg Erlangen, Germany Joschua Conrad [email protected] Universität Ulm Ulm, Germany Maurits Ortmanns [email protected] Universität Ulm Ulm, Germany Vasileios Belagiannis [email protected] Fr |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-12977