Nature Machine Intelligence最新文献

{"title":"Towards understanding and monitorability of reasoning LLMs","authors":"","doi":"10.1038/s42256-026-01333-6","DOIUrl":"10.1038/s42256-026-01333-6","url":null,"abstract":"Frontier LLMs are trained to solve problems in a step-by-step way, generating reasoning text in human-understandable language. However, these reasoning tokens are not accurate explanations of LLMs’ internal computations, raising challenges for monitorability.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1507-1507"},"PeriodicalIF":29.8,"publicationDate":"2026-10-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01333-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149219602","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Reusability report: multi-sample omics network inference operates in distinct robustness regimes defined by prior knowledge and data availability","authors":"Xabier Martínez-de-Morentin, Sumeer A. Khan, Itziar Cenzano, Robert Lehmann, Alberto Maillo, Vincenzo Lagani, Jesper Tegner, David Gomez-Cabrero","doi":"10.1038/s42256-026-01311-y","DOIUrl":"10.1038/s42256-026-01311-y","url":null,"abstract":"Knowledge-constrained network inference enables the reconstruction of regulatory programmes from high-dimensional biological data, yet the conditions under which shared and context-specific structures can be reliably identified remain poorly understood. Pablo Rodriguez-Mier et al. recently introduced CORNETO, a framework for multi-sample constrained inference. Here we reproduce, evaluate and extend CORNETO in terms of robustness, identifiability and bias–variance trade-offs. Using dynamic haematopoiesis and stable cortical datasets, we systematically assess four network reconstruction paradigms while preserving the original method-specific evaluation procedures and benchmark definitions. We show that multi-sample inference generally improves robustness and reproducibility, although gains in predictive performance depend on biological context and the completeness of the prior network. To enhance stability, we introduce bootstrap aggregation and selective re-expansion strategies that improve consensus robustness and recovery while preserving context-specific signals without modifying the underlying optimization scheme or sacrificing generality. Across datasets and methods, our results reveal distinct inference regimes governed by sample availability, prior knowledge and regularization. Together, these findings position network reconstruction as a regime-dependent learning problem and provide practical principles for robust multi-sample biological network inference. This work studies the reusability of CORNETO, a previously developed framework for learning biological networks from multi-omics data. Robust network inference is found to depend on data availability, prior knowledge, regularization and biological context.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1616-1631"},"PeriodicalIF":29.8,"publicationDate":"2026-10-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01311-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149219603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Large-scale iontronic skins with neural-hierarchical processing architecture","authors":"Zhibin Li,Bingxu Da,Junli Shi,Taihong Wang,Wenlong Si,Bowen Li,Hanxiang He,Zelong Liao,Tianyu Wu,Xudong Han,Longyi Bai,He Kong,Chaoyang Song,Chuan Fei Guo","doi":"10.1038/s42256-026-01325-6","DOIUrl":"https://doi.org/10.1038/s42256-026-01325-6","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"56 1","pages":""},"PeriodicalIF":23.8,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149209404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An all-in-one foundation model for multimodal medical image restoration and enhancement","authors":"Luyang Cao, Zhijian Wu, Jian Zhang, Lei Qi, Yang Gao, Hairong Zheng, Yefeng Zheng, Yinghuan Shi","doi":"10.1038/s42256-026-01310-z","DOIUrl":"10.1038/s42256-026-01310-z","url":null,"abstract":"Medical image quality is essential for reliable clinical diagnosis and downstream artificial intelligence analysis, yet routine images are often degraded by noise, motion artefacts, low contrast, low-dose acquisition or modality-specific artefacts. Here we present AMIFound, an all-in-one foundation model for medical image enhancement and restoration across computed tomography, magnetic resonance imaging, positron emission tomography, endoscopy, ultrasound, fundus imaging and X-ray. AMIFound is trained and evaluated on 3.76 million paired high-quality and degraded images spanning 7 representative degradation patterns. A shared encoder captures modality and degradation cues, while a dynamically routed expert restoration pool adapts the enhancement pathway for each input. Across 7 modalities, AMIFound outperforms 38 task-specific and generalist baselines on full-reference image quality metrics, clinician assessment and robustness tests under combined degradations as well as on naturally degraded clinical images. Enhanced images further improve 44 disease detection tasks, tumour stage prediction and survival risk stratification in non-small-cell lung cancer, including improved individual-level prognostic ranking. These results support AMIFound as a domain-specific foundation model for improving low-quality clinical images and strengthening downstream medical artificial intelligence workflows while preserving clinically relevant information. Cao et al. present AMIFound, a foundation model that unifies medical image restoration across seven imaging modalities, delivering robust enhancement across diverse degradations and preserving clinically relevant information for diagnosis.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1596-1615"},"PeriodicalIF":29.8,"publicationDate":"2026-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149209405","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Emotional optimization by newsroom AI needs behavioural evaluation","authors":"Bence Bago, Jean-François Bonnefon","doi":"10.1038/s42256-026-01315-8","DOIUrl":"10.1038/s42256-026-01315-8","url":null,"abstract":"When newsroom artificial intelligence (AI) engages in emotional packaging, it changes how a story feels, shaping whether people approach, trust and share it. As these systems are designed and deployed, AI and behavioural researchers should work together to test whether emotional packaging helps accurate information reach readers, and whether it deepens division.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1508-1511"},"PeriodicalIF":29.8,"publicationDate":"2026-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149090833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Characterizing the role of reasoning tokens in large language models as state over tokens","authors":"Mosh Levy, Zohar Elyoseph, Shauli Ravfogel, Yoav Goldberg","doi":"10.1038/s42256-026-01303-y","DOIUrl":"10.1038/s42256-026-01303-y","url":null,"abstract":"Large language models increasingly rely on generating reasoning tokens that appear to describe a reasoning process performed by the model to arrive at the answer. However, empirical evidence reveals that when read as text, these tokens are not a faithful explanation of the model’s actual reasoning process. To address this gap between appearance and function, we introduce the State over Tokens (SoT) conceptual framework. SoT reframes reasoning tokens not as a linguistic narrative, but as an externalized computational state—the mechanism that persists the computation across the separate generation cycles. The SoT framing explains how the tokens can drive correct reasoning without being a faithful explanation when read as text and surfaces previously overlooked research questions on these tokens. We discuss how it dispels two common misconceptions: that the text records the complete process, and that the model uses the words with the meanings a human reader ascribes to them. We argue that this framing offers a fruitful way to think about these tokens, one that captures their mechanics accurately while remaining simple and intuitive, and that adopting it opens research directions. To better understand the mechanism of large-language-model reasoning, research should move beyond reading the tokens as text and also focus on decoding them as state. When large language models ‘reason’ about a problem, they might still solve it differently from what their reasoning transcript suggest. In this Perspective, Levy et al. describe how one can start investigating the role of these reasoning tokens.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1515-1522"},"PeriodicalIF":29.8,"publicationDate":"2026-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149090836","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Shifting from knowledge retrieval to evidence exploration and synthesis","authors":"Shruti Shikhare, Jake Cohen-Setton, Krishna C. Bulusu","doi":"10.1038/s42256-026-01313-w","DOIUrl":"10.1038/s42256-026-01313-w","url":null,"abstract":"Biomedical discovery has entered an era in which the limiting resource is no longer data, but our ability to integrate and interpret evidence. DeepEvidence, a new deep research agent, goes beyond retrieving facts and constructs explicit representations of scientific evidence.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1512-1514"},"PeriodicalIF":29.8,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149090838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Large language models discover complementary heuristics for combinatorial optimization","authors":"Huatian Gong, Shuaian Wang, Dongping Song, Jiuh-Biing Sheu, Ran Yan","doi":"10.1038/s42256-026-01307-8","DOIUrl":"10.1038/s42256-026-01307-8","url":null,"abstract":"Combinatorial optimization (CO) underpins decision-making across science, engineering and industry, especially manufacturing, logistics, healthcare operations and energy management. However, designing effective heuristics for each new problem variant requires deep domain knowledge and months of expert iteration. Directly prompting large language models (LLMs) can generate executable code, but single-pass synthesis of complete heuristics routinely fails on new CO problems. Here we introduce LLM-driven Algorithm Construction via Complementary Evolution (LACE), a framework that decomposes algorithm design into an input schema, output schema, tool library and heuristic portfolio (the $${mathcal{I}}$$ – $${mathcal{O}}$$ $${mathcal{T}}$$ $${mathcal{H}}$$ interface) and refines the portfolio through time-constrained complementary evolution. The interface specifies a verified problem contract under which the LLM operates, directing model capacity to high-level algorithmic reasoning rather than problem-specific implementation. Complementary evolution then iteratively generates and selects specialist heuristics capable of solving heterogeneous instances within strict runtime budgets. Across 36 classical CO-Bench problems, LACE attains an average score of 0.945, against 0.870 for the strongest existing LLM-based method and 0.571 for direct LLM prompting without any framework, locating the gain in the framework rather than the model. On four structurally new problems, LACE reaches 0.97–0.99, while five existing LLM-based baselines fail to produce any feasible algorithm. These results demonstrate that effective algorithm discovery for such problems can be automated to a substantial degree. Huatian Gong and colleagues developed LACE, a large language model-based framework that designs optimization algorithms. It builds a verified problem contract, then evolves a portfolio of complementary heuristics that together solve problems that no single method can handle.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1569-1580"},"PeriodicalIF":29.8,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149087645","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards foundation-style models for energy-frontier heterogeneous neutrino detectors via self-supervised pretraining","authors":"Saúl Alonso-Monsalve, Fabio Cufino, Umut Kose, Anna Mascellani, André Rubbia","doi":"10.1038/s42256-026-01309-6","DOIUrl":"10.1038/s42256-026-01309-6","url":null,"abstract":"Accelerator-based neutrino physics is entering an energy-frontier regime in which interactions reach the tera-electronvolt scale and produce exceptionally dense, overlapping detector signatures. Event interpretation is impractical for conventional reconstruction and challenging for supervised models trained from scratch, particularly when labelled data are scarce and analyses span diverse objectives. Here we present a sparse Vision Transformer framework for learning reusable representations from heterogeneous detector data. Self-supervised pretraining combines masked-autoencoder reconstruction with relational voxel-level objectives for hierarchy, ghost and particle identification, followed by joint fine-tuning across classification and regression tasks. On simulated events from the proposed FASER calorimeter (FASERCAL) concept, pretraining improves neutrino-flavour and charm-quark identification, momentum regression and vertex reconstruction, with relational objectives providing further gains in topologically complex channels. Attribution and representation diagnostics show a more structured latent space, while detector-subsystem ablations recover physically plausible channel-dependent roles. With roughly 103 labelled events, the pretrained encoder matches the flavour-classification performance of scratch training with an order of magnitude more data. Cross-domain transfer is observed on public plastic-scintillator and liquid-argon benchmarks. A matched alternative-generator stress test shows stable flavour and kinematic performance while exposing charm tagging as generator sensitive. These results support self-supervised multimodal pretraining as a route towards reusable detector representations. We use ‘foundation-style’ in this restricted sense, rather than claiming a completed general-purpose detector foundation model. Alonso-Monsalve et al. demonstrate that self-supervised pretraining helps deep learning models to interpret complex neutrino detector events, improving classification, reconstruction and data efficiency while enabling transfer across different neutrino detector technologies.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1581-1595"},"PeriodicalIF":29.8,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01309-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149087640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Minute-scale training for microrobot navigation","authors":"Yinghan Sun, Aoji Zhu, Xiang Ji, Yamei Li, Jiachi Zhao, Yun Wang, Li Zhang, Huijun Gao, Lidong Yang","doi":"10.1038/s42256-026-01305-w","DOIUrl":"10.1038/s42256-026-01305-w","url":null,"abstract":"Microrobots hold considerable potential for various applications in which targeted navigation is a basic requirement. Deep reinforcement learning has recently emerged as a powerful paradigm for fully autonomous microrobot navigation. Yet, current deep reinforcement learning-based approaches pay limited attention to learning efficiency and effectiveness, requiring hours to days for model training. Consequently, this impedes both rapid practical deployment and parameter optimization. To address these challenges, we present a learning framework that enables effective microrobot navigation policies to be trained within minutes. Here in the proposed framework, we develop a fully vectorized simulator with more than 10,000 artificial vascular environments, parallelizing dynamics, ray-casting-based visual feature extraction and feasibility checks across thousands of environments to achieve roughly 190,000 transitions per second. To achieve effectiveness in the fast training, we propose a task-shaping-regularization reward framework. The task-shaping-regularization framework accelerates convergence, improves final performance, reduces action variation by at least 33.7% and increases obstacle clearance by at least 2.1% across all evaluated scenarios. Results show that the proposed learning framework reduces training time to under 10 minutes, while supporting zero-shot deployment across distinct microrobot types and navigation scenarios. Collectively, this framework can substantially shorten the design loop and accelerate the deployment of autonomous microrobots. A vectorized simulator and structured reward framework enable microrobot navigation policies to be trained within minutes and transferred without retraining across robots and environments.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 10","pages":"1523-1536"},"PeriodicalIF":29.8,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149087641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}