Saved in:
Bibliographic Details
Main Authors: He, Haoyang, Rong, Zihua, Zhao, Liangjie, Zhao, Yunjia, Yang, Lan, Zhang, Honggang
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.03297
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917311627132928
author He, Haoyang
Rong, Zihua
Zhao, Liangjie
Zhao, Yunjia
Yang, Lan
Zhang, Honggang
author_facet He, Haoyang
Rong, Zihua
Zhao, Liangjie
Zhao, Yunjia
Yang, Lan
Zhang, Honggang
contents 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 regime through a continual self-evolving loop. Experimental results on multiple challenging mathematical reasoning benchmarks show that TTSR consistently improves reasoning performance and generalizes well across different model backbones and general-domain reasoning tasks. These findings suggest that teacher-mediated self-reflection provides an effective pathway for stable and continual reasoning improvement at test time.
format Preprint
id arxiv_https___arxiv_org_abs_2603_03297
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TTSR: Test-Time Self-Reflection for Continual Reasoning Improvement
He, Haoyang
Rong, Zihua
Zhao, Liangjie
Zhao, Yunjia
Yang, Lan
Zhang, Honggang
Computation and Language
Artificial Intelligence
Machine Learning
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 regime through a continual self-evolving loop. Experimental results on multiple challenging mathematical reasoning benchmarks show that TTSR consistently improves reasoning performance and generalizes well across different model backbones and general-domain reasoning tasks. These findings suggest that teacher-mediated self-reflection provides an effective pathway for stable and continual reasoning improvement at test time.
title TTSR: Test-Time Self-Reflection for Continual Reasoning Improvement
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2603.03297