Trees to Flows and Back: Unifying Decision Trees and Diffusion Models
Sai Niranjan Ramachandran, Suvrit Sra · 2026
Read Trees to Flows and Back: Unifying Decision Trees and Diffusion Models by Sai Niranjan Ramachandran, Suvrit Sra (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-2605-00414.txt.
Introduction
Sai Niranjan Ramachandran and Suvrit Sra unify two model families that look nothing alike: decision trees, discrete and hierarchical, and diffusion models, continuous and dynamic. They establish a precise mathematical correspondence between hierarchical trees and diffusion processes in suitable limiting regimes. Underneath both, they find one shared optimization principle, which they call Global Trajectory Score Matching.
An idealized form of gradient boosting turns out to be asymptotically optimal for that principle, which ties a workhorse of tabular learning to the theory. Two practical offshoots follow. One generates tabular data with higher fidelity at half the computing cost of rivals. The other, a distillation method, transfers a tree's hierarchical decision logic into a neural network, landing within a whisker of the teacher's performance on many benchmarks. The paper's value is mainly conceptual: two literatures turn out to share a skeleton.
Catalog record
| CREATOR | Sai Niranjan Ramachandran, Suvrit Sra |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 216 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.00414