Efficient Test-Time Scaling via Temporal Reasoning Aggregation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing dynamic early-exit methods rely on single-step confidence signals . existing approaches are unreliable for detecting reasoning convergence in multi-step settings . |
| Approach: | They propose a training-free framework for efficient test-time scaling that determines when to terminate reasoning based on temporal aggregation of multi-step evidence rather than instantaneous signals. |
| Outcome: | Experiments show that TRACE reduces reasoning token usage by 25% on average while maintaining accuracy within 1–2% of full-length reasoning. |
Similar Papers
Thinking Long, but Short: Stable Sequential Test-Time Scaling for Large Reasoning Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Inducing models to think for longer can increase accuracy, but as the length of reasoning is further extended, it has also been shown to result in accuracy degradation and model instability. |
| Approach: | They propose a sequential test-time scaling method which induces models to think for longer, but which also generates an increasingly long output. |
| Outcome: | The proposed method improves model accuracy significantly over a wide range of induced thoughts, stabilizing the accuracy of sequential scaling, and eliminating the need for reasoning length fine-tuning. |
Hidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to speed up parallel scaling have relied on similarity-based or confidence-based pruning, but these signals do not reliably indicate trace quality. |
| Approach: | They propose a pruning framework that evaluates reasoning steps using hidden states and dynamically prunes unpromising traces during generation. |
| Outcome: | The proposed framework reduces end-to-end inference latency by 45%–70% on average compared to self-consistency while improving reasoning accuracy. |
ARISE: An Adaptive Resolution-Aware Metric for Test-Time Scaling Evaluation in Large Reasoning Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing evaluation methods for test-time scaling are limited. |
| Approach: | They propose an adaptive resolution-aware scaling evaluation metric specifically designed to assess the test-time scaling effectiveness of large reasoning models. |
| Outcome: | The proposed metric provides a reliable and fine-grained measurement of test-time scaling capabilities, revealing significant variations in scaling efficiency across models. |
Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering (2025.acl-short)
Copied to clipboard
| Challenge: | Existing evaluations of test-time scaling assume that a reasoning system should always give an answer to any question provided. |
| Approach: | They propose to increase compute budget at inference time to increase confidence in correct responses by considering settings with non-zero levels of response risk. |
| Outcome: | The proposed model can answer more questions correctly and have higher confidence in correct responses. |
Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models (2025.emnlp-main)
Copied to clipboard
Kaiyan Chang, Yonghao Shi, Chenglong Wang, Hang Zhou, Chi Hu, Xiaoqian Liu, Yingfeng Luo, Yuan Ge, Tong Xiao, JingBo Zhu
| Challenge: | Recent training-based TTS methods, such as continued reinforcement learning, have surged in popularity, while training-free TTS approaches are gradually fading from prominence. |
| Approach: | They propose a fine-grained sequential scaling method guided by process verification that integrates training-free TTS methods with other classical parallel scaling methods at the step level. |
| Outcome: | Experiments on five instruction-tuned large language models (LLMs) show that training-free TTS methods can extend reasoning performance boundaries. |
Test-Time Scaling of Reasoning Models for Machine Translation (2026.eacl-long)
Copied to clipboard
| Challenge: | Using TTS, Reasoning Models (RMs) are able to perform tasks such as math and coding with limited results. |
| Approach: | They evaluate 12 Reasoning Models across a diverse suite of MT benchmarks, examining three scenarios: direct translation, forced-reasoning extrapolation, and post-editing. |
| Outcome: | The proposed approach improves translation quality on three domains, with inconsistent results for general-purpose RMs and performance plateauing. |
Thought calibration: Efficient and confident test-time scaling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for teaching language models to be economical with their token budgets have failed to achieve the desired results. |
| Approach: | They propose to calibrate a language model's growing body of thoughts to determine when new reasoning plateaus. |
| Outcome: | The proposed framework preserves model performance with up to 60% reduction in thinking tokens on in-distribution data, and up to 20% in out-of-difference data. |
The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis (2026.acl-long)
Copied to clipboard
Zihao Wei, Liang Pang, Jiahao Liu, Wenjie Shi, Jingcheng Deng, Shicheng Xu, Zenghao Duan, Jingang Wang, Fei Sun, Huawei Shen, Xueqi Cheng
| Challenge: | Explicit reasoning trajectories increase performance but often trigger overthinking . despite its importance, this study examines how each step of reasoning affects the final outcome . |
| Approach: | They propose a Reasoning Completion Point Detector that detects the RCP by monitoring rank dynamics of termination tokens. |
| Outcome: | The proposed method reduces token usage by up to 44% while preserving accuracy. |
The Best of Both Worlds: Combining Parallel and Sequential Inference Scaling via Aggregation Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Empirical results show that AFT-trained models achieve substantial gains with test-time scaling. |
| Approach: | They introduce a supervised fine-tuning paradigm where models synthesize multiple draft responses into a single, refined answer. |
| Outcome: | Empirical results show that AFT-trained models outperform baseline models while eliminating external guidance. |
Answer Convergence as a Signal for Early Stopping in Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | a systematic study suggests that chain-of-thought prompting is unnecessary for producing correct answers. |
| Approach: | They propose three inference-time strategies to improve model efficiency by boosting end-of-reasoning signals and early stopping . they propose a method that learns when to stop based on internal activations . |
| Outcome: | The proposed methods reduce token usage with little or no accuracy drop on natural questions . the proposed methods also reduce tokens by over 40% on naturalquestions . |