NeFut Logo NeFut
Admin Login

[CS.AI] THOR: A Theta-Gamma Framework for Multi-hop QA

Published at: 2026-07-25 22:00 Last updated: 2026-07-26 07:44
#algorithm #AI #Machine Learning

Abstract

Multi-hop question answering requires retrieving and integrating evidence from multiple contexts. Despite the rapid progress of current research, multi-hop reasoning remains constrained by two persistent limitations:

  1. Attention Decay: The model's focus on the main question degrades as the reasoning chain grows.
  2. Error Accumulation: Mistakes propagate across hops, compounding into final failure.

Inspired by Theta-Gamma hierarchical oscillation, which decouples global planning from local retrieval, THOR enables efficient attention transfer between hops and incorporates a verification and repair mechanism to interrupt error accumulation along incorrect paths. Extensive comparative experiments and specific validation experiments on multi-hop QA benchmarks demonstrate that THOR improves answer accuracy and robustness while mitigating these limitations, showcasing its generalization across different backbones.

Blogger's Review: The THOR framework addresses critical issues in multi-hop QA through its innovative Theta-Gamma mechanism, particularly in managing attention and handling errors. Its effective integration of global and local information shows great potential for advancing multi-hop reasoning technologies and their applications.

Original Source: https://arxiv.org/abs/2607.20459

[h] Back to Home