Revisiting Randomization in Greedy Model Search
Xin Chen a, Jason M. Klusowski a,1, Yan Shuo Tan b, Chang Yu a · 2025
Read Revisiting Randomization in Greedy Model Search by Xin Chen a, Jason M. Klusowski a,1, Yan Shuo Tan b, Chang Yu a (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-2506-15643.txt.
Description
Xin Chen and co-authors revisit randomness in greedy model building in a 2025 paper. Random forests sample a subset of features at each step, a trick long treated as mere variance control, like shrinkage. That story rests on smooth fitting methods. The team instead studies greedy forward selection, which acts more like a decision tree's adaptive search.
Under a clean design, they prove sampling features can cut both bias and variance at once. Training error and model size can even move non-monotonically as the sampling rate changes, breaking the shrinkage analogy. In the limit, the method reweights least-squares coefficients by rank, with weights following a logistic curve. Randomness inside a greedy search, it turns out, does different work than in smooth fitting.
Catalog record
| CREATOR | Xin Chen a, Jason M. Klusowski a,1, Yan Shuo Tan b, Chang Yu a |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 118 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-15643