On Generation in Metric Spaces
Jiaxun Li, Vinod Raman, Ambuj Tewari · 2026
Read On Generation in Metric Spaces by Jiaxun Li, Vinod Raman, Ambuj Tewari (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-2602-07710.txt.
Description
Machine learning theory often asks when a system can keep producing genuinely new examples forever, a question previously studied mostly for countable sets. Jiaxun Li and two co-authors move the problem into metric spaces, where novelty means staying a certain distance away from past examples.
They define a scale-sensitive closure dimension that characterizes when uniform generation, non-uniform generation, and generation in the limit are possible, allowing different novelty thresholds for the adversary and the generator. In well-behaved doubling spaces generation stays stable across scales, but in general spaces, even the Hilbert space of square-summable sequences, it can fail abruptly. The paper sharpens the boundary between learnable and unlearnable generation.
Catalog record
| CREATOR | Jiaxun Li, Vinod Raman, Ambuj Tewari |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 116 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07710