mlsanityData to 14 Oct 2026

Learning to Stop Overthinking at Test Time

arXiv 2502.10954 ↗First posted 16 Feb 2025cs.CV cs.AI cs.LGcited by 4

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Test time scaling is currently one of the most active research areas that shows promise after training time scaling has reached its limits. Deep-thinking (DT) models are a class of recurrent models that can perform easy-to-hard generalization by assigning more compute to harder test samples. However, due to their inability to determine the complexity of a test sample, DT models have to use a large amount of computation for both easy and hard test samples. Excessive test time computation is wasteful and can cause the ``overthinking'' problem where more test time computation leads to worse results. In this paper, we introduce a test time training method for determining the optimal amount of computation needed for each sample during test time. We also propose Conv-LiGRU, a novel recurrent architecture for efficient and robust visual reasoning. Extensive experiments demonstrate that Conv-LiGRU is more stable than DT, effectively mitigates the ``overthinking'' phenomenon, and achieves superior accuracy.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 29 of its 34 references that are in the map. First posted on arXiv 16 Feb 2025.

Builds on

29 of 34 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025s1: Simple test-time scaling31 Jan
  2. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May
  3. 2022End-to-end Algorithm Synthesis with Recurrent Networks: Logical Extrapolation Without Overthinking11 Feb

Other references · 26

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  3. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  6. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,234
  7. 2024WorldSimBench: Towards Video Generation Models as World Simulatorscited by 881
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 797
  9. 2023VILA: On Pre-training for Visual Language Modelscited by 692
  10. 2016Adaptive Computation Time for Recurrent Neural Networkscited by 506
  11. 2023Resurrecting Recurrent Neural Networks for Long Sequencescited by 408
  12. 2024Learning to (Learn at Test Time): RNNs with Expressive Hidden Statescited by 355
14 more
  1. 2024xLSTM: Extended Long Short-Term Memorycited by 327
  2. 2024Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Modelscited by 242
  3. 2022Test-Time Training with Masked Autoencoderscited by 212
  4. 2015Delving Deeper into Convolutional Networks for Learning Video Representationscited by 148
  5. 2025LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMscited by 147
  6. 2021Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networkscited by 117
  7. 2024Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Modelscited by 116
  8. 2018Benchmarking Neural Network Robustness to Common Corruptions and Surface Variationscited by 99
  9. 2024Beyond A*: Better Planning with Transformers via Search Dynamics Bootstrappingcited by 90
  10. 2018Light Gated Recurrent Units for Speech Recognitioncited by 48
  11. 2020Differentiable Adaptive Computation Time for Visual Reasoningcited by 12
  12. 2013Understanding Deep Architectures using a Recursive Convolutional Networkcited by 11
  13. 2023LaRS: Latent Reasoning Skills for Chain-of-Thought Reasoningcited by 8
  14. 2023Adaptive recurrent vision performs zero-shot computation scaling to unseen difficulty levelscited by 6

Built on by

Cited by year

Citing papers per year: 2025: 4.

4 papers.

  1. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 384
  2. 2025Retro-Search: Exploring Untaken Paths for Deeper and Efficient Reasoningcited by 34
  3. 2025Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Trainingcited by 16
  4. 2025InsertRank: LLMs can reason over BM25 scores to Improve Listwise Rerankingcited by 4

Similar, not cited

Close in wording, no citation either way

  1. 2025Think Twice: Enhancing LLM Reasoning by Scaling Multi-round Test-time Thinking
  2. 2025Extending Test-Time Scaling: A 3D Perspective with Context, Batch, and Turn
  3. 2026VisRef: Visual Refocusing while Thinking Improves Test-Time Scaling in Multi-Modal Large Reasoning Models
  4. 2026Beyond Model Scaling: Test-Time Intervention for Efficient Deep Reasoning
  5. 2025Iterative Deepening Sampling as Efficient Test-Time Scaling
  6. 2026Thinking Long, but Short: Stable Sequential Test-Time Scaling for Large Reasoning Models
  7. 2025Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning
  8. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs