Explorations of the Softmax Space: Knowing When the Neural Network Doesn’t Know
Daniel Sikar, Artur d’Avila Garcez, Tillman Weyde · 2025
Read Explorations of the Softmax Space: Knowing When the Neural Network Doesn’t Know by Daniel Sikar, Artur d’Avila Garcez, Tillman Weyde (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-2502-00456.txt.
Description
Daniel Sikar, Artur Garcez, and Tillman Weyde give nets a way to say not known when a call rests on shaky ground, in a 2025 paper. Trust in auto calls grows key as nets spread to grave tasks. A high truth net can still make some calls that should go to a human hand.
Last layer score packs bunch by class, and mean packs of right calls mark each class core. A gap cap per class, set by the near wrong call, flags when to hold back. Tests on hand digit shots and broad shot sets show the rule holds across nets. Sight nets and mixed nets both take the hold rule well.
Catalog record
| CREATOR | Daniel Sikar, Artur d’Avila Garcez, Tillman Weyde |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-00456