Nature Machine Intelligence最新文献

{"title":"Reusability report: Exploring the utility and extensibility of an integrated modelling framework for liquid electrolyte design","authors":"Genming Lai, Juntao Zhao, Zekai Liu, Ruiqi Zhang, Hanming Li, Qiliang Zhang, Fangchao Rong, Chi Fang, Qinghua Liu, Yunxing Zuo, Bo Xu, Jiaxin Zheng, Chuying Ouyang","doi":"10.1038/s42256-026-01277-x","DOIUrl":"10.1038/s42256-026-01277-x","url":null,"abstract":"Despite rapid advances in machine learning-driven electrolyte design, the robustness, transferability and applicability of existing models remain insufficiently explored. Yang et al. recently introduced a unified machine learning framework for electrolyte formulation design. The framework adopts a physics-informed architecture that explicitly integrates molecular structural representations with formulation-level compositional information while preserving permutation invariance. By coupling forward property prediction with inverse generation, Yang et al. achieved accurate property prediction and efficient exploration of the formulation space. Here we present a systematic evaluation and multiscale extension of the framework proposed by Yang et al. We assess its robustness and reproducibility through rigorous benchmarking. We also reveal how training data size and compositional heterogeneity govern its applicability by quantifying its sensitivity to data distribution. Furthermore, we demonstrate cross-system transferability through zero-shot and few-shot learning across diverse operational regimes and novel electrolyte compositions. A multiscale extension of the framework has been implemented to encompass diverse targets, ranging from fundamental physical properties to electronic energy boundaries and battery Coulombic efficiency, where it substantially outperforms standard baselines. Overall, this work reveals both the potential and limitations of the framework proposed by Yang et al. across different systems and properties, and it establishes a reference for the reliable application and broader adoption of artificial intelligence-driven electrolyte design. Lai et al. extend and evaluate a unified framework for liquid electrolyte design, showing how data size and composition affect robustness, and demonstrating improved cross-system transferability and multiscale performance over baselines.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 8","pages":"1251-1260"},"PeriodicalIF":29.8,"publicationDate":"2026-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148612648","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":"Classifying multipartite continuous-variable entanglement structures through data-augmented neural networks","authors":"Xiaoting Gao, Mingsheng Tian, Feng-Xiao Sun, Ya-Dong Wu, Yu Xiang, Qiongyi He","doi":"10.1038/s42256-026-01284-y","DOIUrl":"10.1038/s42256-026-01284-y","url":null,"abstract":"Neural networks have emerged as a promising paradigm for quantum information processing, yet they confront the challenge of generating training datasets with sufficient size and rich diversity, which is particularly acute when dealing with multipartite quantum systems. For instance, in the task of classifying different structures of multipartite entanglement in continuous-variable systems, it is necessary to simulate a large number of infinite-dimensional state data that can cover as many types of non-Gaussian states as possible. Here we develop a data-augmented neural network to address this task with homodyne measurement data. A quantum data augmentation method based on classical data processing techniques and quantum physical principles is proposed to efficiently enhance network performance. By testing on randomly generated tripartite and quadripartite states, we demonstrate that the network can infer the entanglement structure among the various partitions, and the accuracies are substantially improved with data augmentation. Our approach allows us to further extend the use of data-driven machine learning techniques to more complex tasks of learning quantum systems encoded in a large Hilbert space. Gao et al. introduce a quantum data augmentation method to enable neural networks to classify multipartite entanglement structures in infinite-dimensional systems, substantially improving accuracy and reducing the data acquisition costs that typically limit training.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 8","pages":"1302-1311"},"PeriodicalIF":29.8,"publicationDate":"2026-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01284-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148612647","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":"Capable language models can outgrow the benefits of collaboration","authors":"Yubin Kim, Ken Gu, Chanwoo Park, Chunjong Park, Samuel Schmidgall, A. Ali Heydari, Yao Yan, Zhihan Zhang, Yuchen Zhuang, Yun Liu, Mark Malhotra, Paul Pu Liang, Hae Won Park, Yuzhe Yang, Xuhai Xu, Yilun Du, Shwetak Patel, Tim Althoff, Daniel McDuff, Xin Liu","doi":"10.1038/s42256-026-01268-y","DOIUrl":"10.1038/s42256-026-01268-y","url":null,"abstract":"Agents, language model-based systems that can reason, plan and act with tools to accomplish tasks, are widely deployed, yet it remains unclear when multi-agent coordination outperforms a strong single agent. Here we conduct a controlled experiment that holds task prompts, tools and compute budgets constant while varying only coordination structure and model capability. Across 260 configurations spanning six benchmarks, five architectures and three LLM families, we derive a predictive model using empirical coordination metrics. Across benchmarks, single-agent baseline performance emerges as the most robust predictor of whether coordination improves or decreases performance. In particular, we identify an empirical capability-saturation threshold beyond which additional agents are unlikely to improve performance. This threshold correctly predicts the effect of multi-agent coordination on performance in 94% of validation configurations on SWE-bench Verified and Terminal-Bench. We therefore interpret this threshold as a practical selection rule rather than a universal scaling principle. A second effect, baseline-scaled error amplification, survives cluster-robust inference (Probust = 0.030) and supports the failure-mode taxonomy. The fitted model achieves cross-validated R2 = 0.373 (0.413 with a task-grounded capability metric) and selects the best architecture in 87% of held-out configurations. These results provide a quantitative framework for within-domain architecture selection and for estimating when multi-agent coordination is likely to improve performance or add overhead. A controlled study of large language model agents across 260 configurations shows when multi-agent collaboration helps or hurts performance, and introduces a predictive model that selects the best architecture in 87% of held-out within-domain configurations.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1157-1172"},"PeriodicalIF":29.8,"publicationDate":"2026-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01268-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148587358","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":"Thinking and rethinking data AI readiness","authors":"","doi":"10.1038/s42256-026-01288-8","DOIUrl":"10.1038/s42256-026-01288-8","url":null,"abstract":"Training machine learning models on high-quality biological datasets can quickly produce abundant results. But as datasets often evolve over time, further work is required to update models and maintain alignment between models and datasets.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1013-1013"},"PeriodicalIF":29.8,"publicationDate":"2026-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01288-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148617411","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":"Neural sampling from cognitive maps enables goal-directed imagination and planning","authors":"Hui Lin, Yukun Yang, Rong Zhao, Giovanni Pezzulo, Wolfgang Maass","doi":"10.1038/s42256-026-01254-4","DOIUrl":"10.1038/s42256-026-01254-4","url":null,"abstract":"Artificial intelligence systems are becoming more intelligent, but at a very high cost in terms of energy consumption and training requirements. By contrast, our brains only require 20 W of energy, they learn online and they can instantly adjust to changing contingencies. This begs the question what data structures, algorithms and learning methods enable brains to achieve that, and whether these can be ported into artificial devices. We are addressing this question for a core feature of intelligence: the capacity to plan and solve problems, including new problems that involve states that were never encountered before. Here we examine three tools that brains are likely to use for achieving that: cognitive maps, stochastic computing and compositional coding. We integrate these tools into a transparent neural network model, and demonstrate its power for flexible planning and problem-solving. Importantly, this approach is suitable for implementation by in-memory computing and other energy-efficient neuromorphic hardware. In particular, it only requires self-supervised local synaptic plasticity that is suited for on-chip learning. Hence, a core feature of brain intelligence—the capacity to generate solutions to problems that were never encountered before—does not require deep neural networks or large language models, and can be implemented in energy-efficient edge devices. Lin et al. introduce a brain-inspired generative model that provides two key features of intelligence: planning and problem-solving. It uses cognitive maps, stochastic computing and compositional coding, and requires only local synaptic plasticity.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1045-1065"},"PeriodicalIF":29.8,"publicationDate":"2026-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13395624/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148585544","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":"A neural network model of free recall learns multiple memory strategies","authors":"Moufan Li, Kristopher T. Jensen, Qiong Zhang, Qihong Lu, Marcelo G. Mattar","doi":"10.1038/s42256-026-01274-0","DOIUrl":"10.1038/s42256-026-01274-0","url":null,"abstract":"Humans exhibit structured patterns of memory recall, including a tendency to recall more recent information and to recall events in the same order they were experienced. Classic computational models explain these patterns by positing that memories incorporate the ongoing ‘temporal context’, formed by smoothly integrating the stimulus history. However, it is unclear whether a single mechanism can account for the full repertoire of human memory strategies, and if this posited mechanism is optimal for recall performance. For example, human memory experts widely apply the ‘memory palace’ strategy, which is empirically better but not captured by temporal context models. Here we show that neural networks optimized for free recall develop diverse retrieval strategies, with only some of them resembling temporal context models. The best-performing models discovered a stimulus-invariant index code that emphasizes the studied position of each list item, instead of its temporal context. This creates a stable scaffold for forward recall akin to the memory palace technique. This index code was more likely to emerge when networks were (1) encouraged to recall all studied items rather than prioritizing only the next few items, and (2) prevented from relying on recency. Our findings demonstrate that human-like recall patterns can arise from multiple distinct computational mechanisms, and that sequential retrieval using item index is an optimal strategy that explains expert-level recall performance. Li et al. show that recurrent neural networks optimized for free recall discover diverse, human-like memory strategies beyond classical temporal context models, with top models using an index-based mechanism resembling the memory palace technique.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 8","pages":"1238-1250"},"PeriodicalIF":29.8,"publicationDate":"2026-07-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148587338","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":"Enabling local neural operators to perform equation-free system-level analysis","authors":"Gianluca Fabiani, Hannes Vandecasteele, Somdatta Goswami, Constantinos Siettos, Ioannis G. Kevrekidis","doi":"10.1038/s42256-026-01265-1","DOIUrl":"10.1038/s42256-026-01265-1","url":null,"abstract":"Neural operators (NOs) offer a powerful computational framework to learn complex spatiotemporal dynamics as mappings between infinite-dimensional function spaces. Still, NOs have primarily been used as surrogates for brute-force temporal simulations and predictions. Their potential for systematic, system-level numerical analysis, such as fixed-point, stability and bifurcation analysis—crucial for predicting irreversible transitions in real-world phenomena—remains largely unexplored. Here, motivated by the equation-free approach, we develop a framework that integrates (local) NOs with iterative numerical analysis methods in the Krylov subspace. This approach expands their potential beyond simulation to efficiently analyse large-scale dynamical systems and tackle fundamental challenges in computer-assisted modelling and numerical analysis of complex systems. Furthermore, we demonstrate the utility of learning local in space–time NOs, which, combined with multiscale equation-free schemes—such as projective integration, Gap-Tooth and Patch Dynamics—accelerate system-level computations, improve the conditioning of Krylov solvers and reduce memory requirements, enabling efficient multiscale analysis of complex spatiotemporal dynamics. We illustrate our framework via three nonlinear partial differential equations (PDE) benchmarks: the one-dimensional Allen–Cahn equation, which undergoes multiple concatenated pitchfork bifurcations, the Liouville–Bratu–Gelfand PDE, which features a saddle-node tipping point, and the FitzHugh–Nagumo model, consisting of two coupled PDEs that exhibit both Hopf and saddle-node bifurcations. Moving beyond brute-force simulations, local neural operators—combined with equation-free methods and Krylov subspace techniques—enable system-level stability and bifurcation analysis of complex spatiotemporal systems directly from data.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1127-1141"},"PeriodicalIF":29.8,"publicationDate":"2026-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148459345","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":"Realigning AI technology towards the Sustainable Development Goals","authors":"Rundong Feng, Ziming Li, Bojie Fu, Huadong Guo, Jianguo Liu, Yoshua Bengio","doi":"10.1038/s42256-026-01271-3","DOIUrl":"10.1038/s42256-026-01271-3","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1014-1016"},"PeriodicalIF":29.8,"publicationDate":"2026-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148432464","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":"A unifying framework from neural superposition to sparse interpretable codes","authors":"David Klindt, Charles O’Neill, Patrik Reizinger, Harald Maurer, Nina Miolane","doi":"10.1038/s42256-026-01259-z","DOIUrl":"10.1038/s42256-026-01259-z","url":null,"abstract":"Understanding how information is represented in neural networks is a fundamental challenge in both neuroscience and artificial intelligence. Despite their nonlinear computations, ample evidence suggests that neural networks encode features in superposition, meaning that systems linearly represent more concepts than they have neurons. This observation opens the door to extracting interpretable representations from otherwise opaque networks, but a principled account of why superposition arises and how it can be exploited has been lacking. Here we synthesize insights from identifiability theory, compressed sensing and quantitative interpretability research to propose a unified perspective on this phenomenon. Our synthesis yields a three-step framework: identifiability theory establishes that neural networks trained for classification recover latent features up to linear mixing, compressed sensing provides guarantees for disentangling these features via sparse coding, and interpretability metrics grounded in behavioural tasks assess whether the extracted features align with human-interpretable concepts. By bridging theoretical neuroscience, representation learning and interpretability research, our framework connects longstanding questions about neural coding in biological systems with modern efforts in artificial intelligence transparency, and highlights open problems at their intersection. Kindt et al. present a unifying framework for superposition in neural networks. Their three-step approach clarifies how latent features can be identified, disentangled and assessed.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1025-1037"},"PeriodicalIF":29.8,"publicationDate":"2026-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148432198","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":"The brain is a diverse place, why not computing?","authors":"Peyton Chandarana, James B. Aimone","doi":"10.1038/s42256-026-01273-1","DOIUrl":"10.1038/s42256-026-01273-1","url":null,"abstract":"The brain’s architecture exhibits diversity across many temporal and spatial scales, yet our computing architectures remain largely homogeneous. Low-powered neuromorphic hardware offers a path towards energy-efficient AI, but could these approaches be improved with heterogeneous computing architectures?","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 7","pages":"1023-1024"},"PeriodicalIF":29.8,"publicationDate":"2026-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148432202","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}