Adaptive Computation

Follow this topic Ask about this topic

Adaptive computation studies how models can spend more processing on difficult or uncertain inputs and skip redundant work on easier ones, improving efficiency without giving up accuracy. Common approaches include early-exit networks, token routing, recurrent and latent reasoning with learned halting, and difficulty- or uncertainty-based compute routers; related methods adapt visual observation and planning budgets. These techniques are tested in settings such as mathematical reasoning, language modeling, multimodal video, edge vision, and robotics, with efficiency measured in FLOPs, tokens, latency, memory, or energy. A central challenge is making predicted compute needs reliable while ensuring savings translate into real speedups, and evaluating gains against strong fixed-compute baselines across changing tasks and hardware.

Latest papers

103 papers23 papers in the last four weeks, up 44% on the four weeks before. 0.1% of all new papers.

Self-Aware Markov Models for Discrete Reasoning
Mar 17, 2026Gregor Kornhardt, Jannis Chemseddine, Christian Wald +1Masked Diffusion Models Adaptive Computation
Technische Universität Berlin, Straße des 17. Juni 136, 10623 Berlin, Germany
Thinking in Latents: Adaptive Anchor Refinement for Implicit Reasoning in LLMs
Mar 16, 2026Disha Sheshanarayana, Rajat Subhra Pal, Manjira Sinha +1Efficient Language Model Reasoning Implicit Reasoning in Language Models
Manipal University Jaipur · TCS Research
CAMD: Coverage-Aware Multimodal Decoding for Efficient Reasoning of Multimodal Large Language Models
Mar 16, 2026Huijie Guo, Jingyao Wang, Lingyu Si +3Efficient Multimodal Inference Multimodal Large Language Models
Institute of Software Chinese Academy of Sciences · University of Chinese Academy of Sciences · Wangxuan Institute of Computer Technology, Peking University
CODA: Difficulty-Aware Compute Allocation for Adaptive Reasoning
Mar 9, 2026Siye Wu, Jian Xie, Yikai Zhang +1Adaptive Reasoning in Language Models Adaptive Computation
♠Fudan University
DOVA: Deliberation-First Multi-Agent Orchestration for Autonomous Research Automation
Mar 4, 2026Aaron Shen, Alfred ShenMulti-Agent Orchestration Autonomous Research Agents
University of California, Berkeley, USA · Amazon Web Services, USA
Architectural Proprioception in State Space Models: Thermodynamic Training Induces Anticipatory Halt Detection
Mar 4, 2026Jay NoonState Space Models Metacognition in Language Models
PonderLM-3: Adaptive Token-Wise Pondering with Differentiable Masking
Mar 2, 2026He Li, Feichen Song, Boyi Zeng +4Adaptive Computation Inference-Time Adaptation of Language Models
LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University · Shanghai AI Laboratory · Shanghai Jiao Tong University +1
Fast-Slow Thinking RM: Efficient Integration of Scalar and Generative Reward Models
Mar 2, 2026Jiayun Wu, Peng Zhang, Yuanyuan Lu +3Reward Modeling Efficient Language Model Inference
Fudan University · Meituan
AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth
Mar 2, 2026Shixiang Song, He Li, Zitong Wang +6Recurrent Language Models Inference-Time Adaptation of Language Models
LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University · Shanghai Innovation Institute · Sun Yat-sen University +2
Ruyi2 Technical Report
Feb 26, 2026Huan Song, Shuyu Tian, Junyi Hao +5LLM Inference Acceleration Adaptive Computation
Institute of Artificial Intelligence (TeleAI), China Telecom
Think $^{2}$ : Grounded Metacognitive Reasoning in Large Language Models
Feb 21, 2026Abraham Paul Elenjical, Vivek Hruday Kavuri, Vasudeva VarmaLanguage Model Self-Assessment Metacognition in Language Models
IIIT Hyderabad
Scaling Latent Reasoning via Looped Language Models
Oct 29, 2025Rui-Jie Zhu, Zixuan Wang, Kai Hua +30Language Model Pretraining Implicit Reasoning in Language Models
ByteDance Seed · UC Santa Cruz · Princeton University +8
Granular-ball computing: an efficient, robust, and interpretable adaptive multi-granularity representation and computation method
Apr 21, 2023Shuyin Xia, Guoyin Wang, Xinbo Gao +2Interpretable ML Adaptive Computation
Chongqing Key Laboratory of Computational Intelligence, Key Laboratory of Cyberspace Big Data Intelligent Security, Ministry of Education, and Key Laboratory of Big Data Intelligent Computing, Chongqing University of Posts and Telecommunications, 400065, Chongqing, China · College of Computer and Information Science, the National Center for Applied Mathematics in Chongqing, Chongqing Normal University, Chongqing 401331, China

Mar 17, 2026J/K move · Enter open · S save

Self-Aware Markov Models for Discrete Reasoning

Gregor Kornhardt, Jannis Chemseddine, Christian Wald, Gabriele Steidl

Technische Universität Berlin, Straße des 17. Juni 136, 10623 Berlin, Germany

Masked Diffusion Models Adaptive Computation LLM Self-Correction

Open paper AskPDF arXiv

Standard masked discrete diffusion models face limitations in reasoning tasks due to their inability to correct their own mistakes on the masking path. Since they rely on a fixed number of denoising steps, they are unable to adjust their computation to the complexity of a given problem. To address these limitations, we introduce a method based on learning a Markov transition kernel that is trained on its own outputs. This design enables tokens to be remasked, allowing the model to correct its previous mistakes. Furthermore, we do not need a fixed time schedule but use a trained stopping criterion. This allows for adaptation of the number of function evaluations to the difficulty of the reasoning problem. Our adaptation adds two lightweight prediction heads, enabling reuse and fine-tuning of existing pretrained models. On the Sudoku-Extreme dataset we clearly outperform other flow based methods with a validity of 95%. For the Countdown-4 we only need in average of 10 steps to solve almost 96% of them correctly, while many problems can be solved already in 2 steps.