Rethinking the Harmonic Loss via Non-Euclidean Distance Layers DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computing methodologies Machine learning Machine learning approaches Neural networks CCS: Computing methodologies Machine learning Learning paradigms Supervised learning CCS: Computing methodologies Artificial intelligence Computer vision CCS: Computing methodologies Artificial intelligence Natural language processing CCS: Theory of computation Design and analysis of algorithms Optimization algorithms and problems
Maxwell Miller-Golub, Collin Coil, Kamil Faber, Marcin Pietron, Panpan Zheng, Pasquale Minervini, Roberto Corizzo · 2026
Read Rethinking the Harmonic Loss via Non-Euclidean Distance Layers DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computing methodologies Machine learning Machine learning approaches Neural networks CCS: Computing methodologies Machine learning Learning paradigms Supervised learning CCS: Computing methodologies Artificial intelligence Computer vision CCS: Computing methodologies Artificial intelligence Natural language processing CCS: Theory of computation Design and analysis of algorithms Optimization algorithms and problems by Maxwell Miller-Golub, Collin Coil, Kamil Faber, Marcin Pietron, Panpan Zheng, Pasquale Minervini, Roberto Corizzo (2026) 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-2603-10225.txt.
Introduction
Maxwell Miller-Golub and coauthors widen the harmonic loss for neural networks beyond plain Euclidean distance. Cross entropy training suffers weak interpretability and unbounded weight growth, while the harmonic alternative stays tied to one geometry. They test many distance metrics inside distance tailored harmonic losses on vision backbones and language models alike. Cosine distance gives the best overall trade, raising accuracy while lowering carbon emissions; other metrics sharpen interpretability.
Catalog record
| CREATOR | Maxwell Miller-Golub, Collin Coil, Kamil Faber, Marcin Pietron, Panpan Zheng, Pasquale Minervini, Roberto Corizzo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 157863 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.10225