Adaptive Multi-Expert Reasoning via Difficulty-Aware Routing and Uncertainty-Guided Aggregation

2026-04-11

AI Summary

This work addresses the instability of large language models in mathematical reasoning across problems of varying difficulty. To mitigate this issue, the authors propose an adaptive multi-expert reasoning framework that coordinates multiple expert models through a difficulty-aware routing mechanism and an uncertainty-guided dynamic sampling strategy. The framework further enhances robustness by integrating a neural verifier with a clustering-based aggregation scheme. Notably, the method operates solely on original training data—without reliance on synthetic examples—and achieves a 75.28% accuracy on the GSM8K benchmark, outperforming most existing 7B-scale models that depend on augmented or synthetic data. This result underscores the framework’s capacity for efficient and robust mathematical reasoning without requiring additional training data.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Natural Language Processing: Safety and Robustness

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large language models (LLMs) demonstrate strong performance in math reasoning benchmarks, but their performance varies inconsistently across problems with varying levels of difficulty. This paper describes Adaptive Multi-Expert Reasoning (AMR), a framework that focuses on problem complexity by reasoning with dynamically adapted strategies. An agile routing system that focuses on problem text predicts problems' difficulty and uncertainty and guides a reconfigurable sampling mechanism to manage the breadth of generation. Three specialized experts create candidate responses, which are modified during multiple correction and finalization phases. A neural verifier assesses the correctness of responses, while a clustering-based aggregation technique identifies the final candidate answer based on a combination of consensus and answer quality. When evaluated on the GSM8K dataset, AMR achieved 75.28% accuracy while only using the original training data. This result outperformed the majority of comparable 7B models that were trained on synthetic data. This showcases that models using difficulty-based routing and uncertainty-driven aggregation are efficient and effective in improving math reasoning models' robustness.

Problem

• math reasoning
• problem difficulty
• large language models
• reasoning robustness

Innovation

• difficulty-aware routing
• uncertainty-guided aggregation
• adaptive multi-expert reasoning
• neural verifier
• clustering-based aggregation