[2603.03297] TTSR: Test-Time Self-Reflection for Continual Reasoning Improvement

Computer Science > Computation and Language arXiv:2603.03297 (cs) [Submitted on 6 Feb 2026] Title:TTSR: Test-Time Self-Reflection for Continual Reasoning Improvement Authors:Haoyang He, Zihua Rong, Liangjie Zhao, Yunjia Zhao, Lan Yang, Honggang Zhang View a PDF of the paper titled TTSR: Test-Time Self-Reflection for Continual Reasoning Improvement, by Haoyang He and Zihua Rong and Liangjie Zhao and Yunjia Zhao and Lan Yang and Honggang Zhang View PDF HTML (experimental) Abstract:Test-time Training enables model adaptation using only test questions and offers a promising paradigm for improving the reasoning ability of large language models (LLMs). However, it faces two major challenges: test questions are often highly difficult, making self-generated pseudo-labels unreliable, and existing methods lack effective mechanisms to adapt to a model's specific reasoning weaknesses, leading to inefficient learning. To address these issues, we propose \textbf{TTSR}, a self-reflective test-time self-evolving training framework. TTSR employs a single pretrained language model that alternates between the roles of a \textit{Student} and a \textit{Teacher} at test time. The Student focuses on solving problems and learning from synthesized variant questions, while the Teacher analyzes the Student's failed reasoning trajectories, summarizes recurring reasoning weaknesses, and synthesizes targeted variant questions accordingly. This process guides the model to improve within a learnable regi...