Related papers: Adaptive Overclocking: Dynamic Control of Thinking Path Length via Real-Time Reasoning Signals

Adaptive Overclocking: Dynamic Control of Thinking Path Length via Real-Time Reasoning Signals

URL: http://arxiv.org/abs/2509.17000v1
Date: Sun, 21 Sep 2025 09:40:27 GMT
Title: Adaptive Overclocking: Dynamic Control of Thinking Path Length via Real-Time Reasoning Signals
Authors: Shuhao Jiang, Songbo Wang, Yang Qiao, Chun Xu, Chaoyang Zheng, Shengyi Zhou, Huanjun Wang, Fangming Li, Cong Zhang, Jiyu Wang,
Abstract summary: We propose Adaptive Overclocking, a method that makes the hyper parameter $alpha$ dynamic and context-aware.<n>Our method adjusts reasoning speed in real time through two complementary signals.<n> Experiments on GSM8K, MATH, and SVAMP show that HAC achieves superior accuracy-latency trade-offs.
Score: 8.264189366042675
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
Abstract: Large Reasoning Models (LRMs) often suffer from computational inefficiency due to overthinking, where a fixed reasoning budget fails to match the varying complexity of tasks. To address this issue, we propose Adaptive Overclocking, a method that makes the overclocking hyperparameter $\alpha$ dynamic and context-aware. Our method adjusts reasoning speed in real time through two complementary signals: (1) token-level model uncertainty for fine-grained step-wise control, and (2) input complexity estimation for informed initialization. We implement this approach with three strategies: Uncertainty-Aware Alpha Scheduling (UA-$\alpha$S), Complexity-Guided Alpha Initialization (CG-$\alpha$I), and a Hybrid Adaptive Control (HAC) that combines both. Experiments on GSM8K, MATH, and SVAMP show that HAC achieves superior accuracy-latency trade-offs, reducing unnecessary computation on simple problems while allocating more resources to challenging ones. By mitigating overthinking, Adaptive Overclocking enhances both efficiency and overall reasoning performance.

Related papers

Quasi-Periodic Gaussian Process Predictive Iterative Learning Control [7.0206339525686055]
Iterative learning control (ILC) improves performance by using information from previous iterations to compensate for expected errors in future iterations.<n>This work incorporates the use of Quasi-Periodic Gaussian Processes (QPGPs) into a predictive ILC framework.<n>We benchmark the method against both standard ILC and conventional GP-based predictive ILC on three tasks.
arXiv Detail & Related papers (2026-02-20T06:10:10Z)
Logit-Entropy Adaptive Stopping Heuristic for Efficient Chain-of-Thought Reasoning [0.0]
Chain-of-Thought (CoT) prompting is a key technique for enabling complex reasoning in large language models.<n>We introduce LEASH: Logit-Entropy Adaptive Stopping Heuristic, a training-free decoding algorithm that adaptively halts rationale generation.
arXiv Detail & Related papers (2025-11-06T18:43:16Z)
DART: Difficulty-Adaptive Reasoning Truncation for Efficient Large Language Models [36.962276192354174]
textbfDART adjusts thinking length according to problem difficulty.<n>textbfTruncation framework learns when to stop thinking''
arXiv Detail & Related papers (2025-11-03T02:41:20Z)
e1: Learning Adaptive Control of Reasoning Effort [88.51897900019485]
Increasing the thinking budget of AI models can significantly improve accuracy, but not all questions warrant the same amount of reasoning.<n>Users may prefer to allocate different amounts of reasoning effort depending on how they value output quality versus latency and cost.<n>We propose Adaptive Effort Control, a self-adaptive reinforcement learning method that trains models to use a user-specified fraction of tokens.
arXiv Detail & Related papers (2025-10-30T23:12:21Z)
A1: Asynchronous Test-Time Scaling via Conformal Prediction [112.54016379556073]
Large language models (LLMs) benefit from test-time scaling, but existing methods face significant challenges.<n>We introduce A1 (Asynchronous Test-Time Scaling), a statistically guaranteed adaptive inference framework that addresses these challenges.<n>A1 achieves a remarkable 56.7x speedup in test-time scaling and a 4.14x improvement in throughput.
arXiv Detail & Related papers (2025-09-18T16:55:09Z)
Controlling Thinking Speed in Reasoning Models [41.72496532709135]
Human cognition operates in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking.<n>In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment.<n>Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance.
arXiv Detail & Related papers (2025-07-04T16:41:06Z)
Fractional Reasoning via Latent Steering Vectors Improves Inference Time Compute [57.16286134405821]
We propose Fractional Reasoning, a framework that enables continuous control over reasoning intensity at inference time.<n>Our method operates by extracting the latent steering vector associated with deeper reasoning and reapplying it with a tunable scaling factor.<n> Experiments on GSM8K, MATH500, and GPQA demonstrate that Fractional Reasoning consistently improves performance across diverse reasoning tasks and models.
arXiv Detail & Related papers (2025-06-18T21:15:59Z)
Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition [95.54406667705999]
Pangu Embedded is an efficient Large Language Model (LLM) reasoner developed on Ascend Neural Processing Units (NPUs)<n>It addresses the significant computational costs and inference latency challenges prevalent in existing reasoning-optimized LLMs.<n>It delivers rapid responses and state-of-the-art reasoning quality within a single, unified model architecture.
arXiv Detail & Related papers (2025-05-28T14:03:02Z)
ARM: Adaptive Reasoning Model [36.53965139929349]
We propose Adaptive Reasoning Model (ARM), a reasoning model capable of adaptively selecting appropriate formats based on the task at hand.<n>Ada-GRPO enables ARM to achieve high token efficiency, reducing tokens by an average of 30%, and up to 70%, while maintaining performance comparable to the model that relies solely on Long CoT.
arXiv Detail & Related papers (2025-05-26T17:38:50Z)
PATS: Process-Level Adaptive Thinking Mode Switching [53.53401063490537]
Current large-language models (LLMs) typically adopt a fixed reasoning strategy, either simple or complex, for all questions, regardless of their difficulty.<n>This neglect of variation in task and reasoning process complexity leads to an imbalance between performance and efficiency.<n>Existing methods attempt to implement training-free fast-slow thinking system switching to handle problems of varying difficulty, but are limited by coarse-grained solution-level strategy adjustments.<n>We propose a novel reasoning paradigm: Process-Level Adaptive Thinking Mode Switching (PATS), which enables LLMs to dynamically adjust their reasoning strategy based on the difficulty of each step, optimizing the balance between
arXiv Detail & Related papers (2025-05-25T17:58:50Z)
Non-Asymptotic Guarantees for Average-Reward Q-Learning with Adaptive Stepsizes [4.169915659794567]
This work presents the first finite-time analysis for the last-iterate convergence of average-reward Q-learning with an asynchronous implementation.<n>A key feature of the algorithm we study is the use of adaptive stepsizes, which serve as local clocks for each state-action pair.
arXiv Detail & Related papers (2025-04-25T23:41:14Z)
Doubly Robust Off-Policy Actor-Critic: Convergence and Optimality [131.45028999325797]
We develop a doubly robust off-policy AC (DR-Off-PAC) for discounted MDP. DR-Off-PAC adopts a single timescale structure, in which both actor and critics are updated simultaneously with constant stepsize. We study the finite-time convergence rate and characterize the sample complexity for DR-Off-PAC to attain an $epsilon$-accurate optimal policy.
arXiv Detail & Related papers (2021-02-23T18:56:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.