CATS: Category-Aware Token-level Steering for Training-Free Redundancy Reduction in Large Reasoning Models

Mengfei Zhang; Zhenglin Wang

AAAI 2026technical0 citations

CATS: Category-Aware Token-level Steering for Training-Free Redundancy Reduction in Large Reasoning Models

Abstract

While Large Reasoning Models (LRMs) exhibit remarkable capabilities in complex tasks, they often suffer from excessive redundancy in their chain-of-thought reasoning. This significantly reduces inference efficiency and increases computational costs. We identify that LRM redundancy is not uniformly homogeneous but can be taxonomized according to whether it is destructive to the final answer: destructive redundancy (e.g., logical drift, hallucination amplification) versus non-destructive redundancy (e.g., repetition, over-elaboration). Moreover, LRM

BibTeX

@inproceedings{aaai2026_catscategoryawar,
  title = {CATS: Category-Aware Token-level Steering for Training-Free Redundancy Reduction in Large Reasoning Models},
  author = {Mengfei Zhang and Zhenglin Wang},
  booktitle = {AAAI 2026},
  year = {2026}
}