Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration

Longxuan Wei, Yubo Zhang, Zijiao Zhang, Zhihu Wang, Shiwan Zhao, Tianyu Huang, Huiting Zhao, Chenfei Liu, Shenao Zhang, Junchi Yan · 2026

Read Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration by Longxuan Wei, Yubo Zhang, Zijiao Zhang, Zhihu Wang, Shiwan Zhao, Tianyu Huang, Huiting Zhao, Chenfei Liu, Shenao Zhang, Junchi Yan (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-2601-15296.txt.

Description

Longxuan Wei, Yubo Zhang, and co-authors guide a language model's search with its own doubt. Their 2026 arXiv paper deals with decoding, the step where a model picks its next words. Current methods either sample at random or repeat the same sampling many times.

The team proposes Entropy-Tree, a tree-based decoding method. Entropy measures how unsure the model is at each step. The search tree branches only where that doubt is genuine. Wasted branches are skipped, and the model's effort goes to the points where the answer is truly in play.

Catalog record

CREATORLongxuan Wei, Yubo Zhang, Zijiao Zhang, Zhihu Wang, Shiwan Zhao, Tianyu Huang, Huiting Zhao, Chenfei Liu, Shenao Zhang, Junchi Yan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-15296

Browse the full library · About the Tower of Babel