Navigating High Dimensional Concept Space with Metalearning
Max Gupta 1 [email protected] 1 Department of Computer Science, Princeton University · 2025
Read Navigating High Dimensional Concept Space with Metalearning by Max Gupta 1 [email protected] 1 Department of Computer Science, Princeton University (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-2508-01948.txt.
Description
Max Gupta asks in a 2025 paper whether meta-learning helps networks grasp abstract concepts from few examples. People learn new rules fast from scant data, a feat machines still chase. His test bed uses logical statements built by a grammar.
He varies both the number of features and the depth of nested rules. Meta-learners handled nested, compositional complexity far better than raw feature counts. Weight and loss-surface checks show feature load roughens the learning path, favoring curvature-aware methods. More adaptation steps improved transfer to harder unseen concepts for one method. The study maps where meta-learning pays and where it stalls.
Catalog record
| CREATOR | Max Gupta 1 [email protected] 1 Department of Computer Science, Princeton University |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-01948