Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration
This addresses the issue of redundant exploration in reasoning tasks for users of large language models, representing an incremental improvement over existing decoding methods.
The paper tackled the problem of inefficient decoding strategies in large language models by proposing Entropy-Tree, a tree-based method that uses entropy to guide branching decisions, resulting in superior accuracy and calibration, such as better pass@k than Multi-chain and improved AUROC for predictive entropy.
Large language models achieve strong reasoning performance, yet existing decoding strategies either explore blindly (random sampling) or redundantly (independent multi-sampling). We propose Entropy-Tree, a tree-based decoding method that exploits entropy as a signal for branching decisions--expanding the search tree only at positions where the model exhibits genuine uncertainty. Entropy-Tree shows superior accuracy and calibration in reasoning tasks: it achieves better pass@k than Multi-chain across multiple models and datasets, and its predictive entropy demonstrates better AUROC compared to several traditional metrics. Entropy-Tree unifies efficient structured exploration and reliable uncertainty estimation within a single decoding procedure.