Reasoning Planning for Language Models

2025-11-01

AI Summary

Existing language model inference path selection relies on redundant candidate generation and neglects computational cost. Method: This paper proposes EPIC—a framework that employs contrastive learning to construct a unified representation space, jointly modeling accuracy and inference overhead for query-adaptive optimal path selection. Theoretically, it provides the first precision analysis of candidate aggregation methods and derives an error bound, enabling a probabilistic-bound regularization mechanism that significantly reduces generation overhead while preserving accuracy. Methodologically, EPIC integrates ensemble-based planning, shared representation learning, and utility-driven optimization. Results: On multiple mathematical reasoning benchmarks, EPIC achieves higher and more stable accuracy with substantially lower computational resources, empirically validating both the optimizability of inference decisions and the effectiveness of co-designing efficiency and accuracy.

Technology Category

• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Planning with Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Selecting an appropriate reasoning method for a given query remains a key challenge in language model generation. Existing approaches typically generate multiple candidate responses and use an aggregation strategy to select the output answer, often assuming that more candidate answers yield higher accuracy. We revisit this assumption through a rigorous theoretical analysis, deriving accuracy bounds for standard aggregation methods under fixed generation distributions and candidate sizes. Building on these insights, we introduce EPIC, an Ensemble Planning with Contrastive learning framework to learn a shared representation space that captures both model reasoning abilities and query-method compatibility. EPIC incorporates our probability bounds as a regularizer in a utility-driven optimization that balances accuracy and computational cost. Experiments on diverse mathematical reasoning tasks show that EPIC consistently selects optimal reasoning methods, improving accuracy while reducing computational overhead. Our code can be found at https://github.com/nguyenngocbaocmt02/EPIC.

Problem

• Selecting optimal reasoning methods for language model queries
• Improving accuracy while reducing computational costs
• Learning query-method compatibility through contrastive representation learning

Innovation

• EPIC framework uses contrastive learning for shared representation
• Probability bounds regularize utility-driven optimization balancing cost
• Selects optimal reasoning methods improving accuracy reducing computation