Nature computational science最新文献_第4页

A new tool for shape and structure optimization of soft materials 软质材料形状和结构优化的新工具。

IF 12

Nature computational science Pub Date : 2025-01-09 DOI: 10.1038/s43588-024-00754-w

引用次数: 0

Efficient scaling of large language models with mixture of experts and 3D analog in-memory computing 高效缩放与专家和三维模拟内存计算混合的大型语言模型。

IF 12

Nature computational science Pub Date : 2025-01-08 DOI: 10.1038/s43588-024-00753-x

Julian Büchel, Athanasios Vasilopoulos, William Andrew Simon, Irem Boybat, HsinYu Tsai, Geoffrey W. Burr, Hernan Castro, Bill Filipiak, Manuel Le Gallo, Abbas Rahimi, Vijay Narayanan, Abu Sebastian

{"title":"Efficient scaling of large language models with mixture of experts and 3D analog in-memory computing","authors":"Julian Büchel, Athanasios Vasilopoulos, William Andrew Simon, Irem Boybat, HsinYu Tsai, Geoffrey W. Burr, Hernan Castro, Bill Filipiak, Manuel Le Gallo, Abbas Rahimi, Vijay Narayanan, Abu Sebastian","doi":"10.1038/s43588-024-00753-x","DOIUrl":"10.1038/s43588-024-00753-x","url":null,"abstract":"Large language models (LLMs), with their remarkable generative capacities, have greatly impacted a range of fields, but they face scalability challenges due to their large parameter counts, which result in high costs for training and inference. The trend of increasing model sizes is exacerbating these challenges, particularly in terms of memory footprint, latency and energy consumption. Here we explore the deployment of ‘mixture of experts’ (MoEs) networks—networks that use conditional computing to keep computational demands low despite having many parameters—on three-dimensional (3D) non-volatile memory (NVM)-based analog in-memory computing (AIMC) hardware. When combined with the MoE architecture, this hardware, utilizing stacked NVM devices arranged in a crossbar array, offers a solution to the parameter-fetching bottleneck typical in traditional models deployed on conventional von-Neumann-based architectures. By simulating the deployment of MoEs on an abstract 3D AIMC system, we demonstrate that, due to their conditional compute mechanism, MoEs are inherently better suited to this hardware than conventional, dense model architectures. Our findings suggest that MoEs, in conjunction with emerging 3D NVM-based AIMC, can substantially reduce the inference costs of state-of-the-art LLMs, making them more accessible and energy-efficient. This study shows a viable pathway to the efficient deployment of state-of-the-art large language models using mixture of experts on 3D analog in-memory computing hardware.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 1","pages":"13-26"},"PeriodicalIF":12.0,"publicationDate":"2025-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142959948","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Decoupled peak property learning for efficient and interpretable electronic circular dichroism spectrum prediction 解耦峰属性学习用于高效和可解释的电子圆二色光谱预测。

IF 12

Nature computational science Pub Date : 2025-01-03 DOI: 10.1038/s43588-024-00757-7

Hao Li, Da Long, Li Yuan, Yu Wang, Yonghong Tian, Xinchang Wang, Fanyang Mo

{"title":"Decoupled peak property learning for efficient and interpretable electronic circular dichroism spectrum prediction","authors":"Hao Li, Da Long, Li Yuan, Yu Wang, Yonghong Tian, Xinchang Wang, Fanyang Mo","doi":"10.1038/s43588-024-00757-7","DOIUrl":"10.1038/s43588-024-00757-7","url":null,"abstract":"Electronic circular dichroism (ECD) spectra contain key information about molecular chirality by discriminating the absolute configurations of chiral molecules, which is crucial in asymmetric organic synthesis and the drug industry. However, existing predictive approaches lack the consideration of ECD spectra owing to the data scarcity and the limited interpretability to achieve trustworthy prediction. Here we establish a large-scale dataset for chiral molecular ECD spectra and propose ECDFormer for accurate and interpretable ECD spectrum prediction. ECDFormer decomposes ECD spectra into peak entities, uses the QFormer architecture to learn peak properties and renders peaks into spectra. Compared with spectrum sequence prediction methods, our decoupled peak prediction approach substantially enhances both accuracy and efficiency, improving the peak symbol accuracy from 37.3% to 72.7% and decreasing the time cost from an average of 4.6 central processing unit hours to 1.5 s. Moreover, ECDFormer demonstrated its ability to capture molecular orbital information directly from spectral data using the explainable peak-decoupling approach. Furthermore, ECDFormer proved to be equally proficient at predicting various types of spectrum, including infrared and mass spectroscopies, highlighting its substantial generalization capabilities. A large-scale electronic circular dichroism spectrum dataset is proposed and the ECDFormer framework is developed to achieve accurate and interpretable ECD spectrum prediction for natural products.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 3","pages":"234-244"},"PeriodicalIF":12.0,"publicationDate":"2025-01-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142928880","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Large language models act as if they are part of a group 大型语言模型的行为就好像它们是一个群体的一部分。

IF 12

Nature computational science Pub Date : 2025-01-02 DOI: 10.1038/s43588-024-00750-0

Germans Savcisens

引用次数: 0

Approaching coupled-cluster accuracy for molecular electronic structures with multi-task learning 基于多任务学习的分子电子结构耦合聚类精度逼近。

IF 12

Nature computational science Pub Date : 2024-12-27 DOI: 10.1038/s43588-024-00747-9

Hao Tang, Brian Xiao, Wenhao He, Pero Subasic, Avetik R. Harutyunyan, Yao Wang, Fang Liu, Haowei Xu, Ju Li

{"title":"Approaching coupled-cluster accuracy for molecular electronic structures with multi-task learning","authors":"Hao Tang, Brian Xiao, Wenhao He, Pero Subasic, Avetik R. Harutyunyan, Yao Wang, Fang Liu, Haowei Xu, Ju Li","doi":"10.1038/s43588-024-00747-9","DOIUrl":"10.1038/s43588-024-00747-9","url":null,"abstract":"Machine learning plays an important role in quantum chemistry, providing fast-to-evaluate predictive models for various properties of molecules; however, most existing machine learning models for molecular electronic properties use density functional theory (DFT) databases as ground truth in training, and their prediction accuracy cannot surpass that of DFT. In this work we developed a unified machine learning method for electronic structures of organic molecules using the gold-standard CCSD(T) calculations as training data. Tested on hydrocarbon molecules, our model outperforms DFT with several widely used hybrid and double-hybrid functionals in terms of both computational cost and prediction accuracy of various quantum chemical properties. We apply the model to aromatic compounds and semiconducting polymers, evaluating both ground- and excited-state properties. The results demonstrate the model’s accuracy and generalization capability to complex systems that cannot be calculated using CCSD(T)-level methods due to scaling. A multi-task deep learning method for molecular electronic structures, called MEHnet, is developed to predict various molecular properties in a unified framework, approaching chemical accuracy while exhibiting local DFT-level computational costs.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 2","pages":"144-154"},"PeriodicalIF":12.0,"publicationDate":"2024-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142900898","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A programmable environment for shape optimization and shapeshifting problems 形状优化和变形问题的可编程环境。

IF 12

Nature computational science Pub Date : 2024-12-27 DOI: 10.1038/s43588-024-00749-7

Chaitanya Joshi, Daniel Hellstein, Cole Wennerholm, Eoghan Downey, Emmett Hamilton, Samuel Hocking, Anca S. Andrei, James H. Adler, Timothy J. Atherton

{"title":"A programmable environment for shape optimization and shapeshifting problems","authors":"Chaitanya Joshi, Daniel Hellstein, Cole Wennerholm, Eoghan Downey, Emmett Hamilton, Samuel Hocking, Anca S. Andrei, James H. Adler, Timothy J. Atherton","doi":"10.1038/s43588-024-00749-7","DOIUrl":"10.1038/s43588-024-00749-7","url":null,"abstract":"Soft materials underpin many domains of science and engineering, including soft robotics, structured fluids, and biological and particulate media. In response to applied mechanical, electromagnetic or chemical stimuli, such materials typically change shape, often dramatically. Predicting their structure is of great interest to facilitate design and mechanistic understanding, and can be cast as an optimization problem where a given energy function describing the physics of the material is minimized with respect to the shape of the domain and additional fields. However, shape-optimization problems are very challenging to solve, and there is a lack of suitable simulation tools that are both readily accessible and general in purpose. Here we present an open-source programmable environment, Morpho, and demonstrate its versatility by showcasing a range of applications from different areas of soft-matter physics: swelling hydrogels, complex fluids that form aspherical droplets, soap films and membranes, and filaments. This study introduces an extensible framework—Morpho—for shape optimization, enabling researchers to predict the structure of soft materials, such as complex fluids, gels, particulate and biological materials.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 2","pages":"170-183"},"PeriodicalIF":12.0,"publicationDate":"2024-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142900892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Leveraging pharmacovigilance data to predict population-scale toxicity profiles of checkpoint inhibitor immunotherapy 利用药物警戒数据来预测检查点抑制剂免疫疗法的人群规模毒性概况。

IF 12

Nature computational science Pub Date : 2024-12-23 DOI: 10.1038/s43588-024-00748-8

Dongxue Yan, Siqi Bao, Zicheng Zhang, Jie Sun, Meng Zhou

{"title":"Leveraging pharmacovigilance data to predict population-scale toxicity profiles of checkpoint inhibitor immunotherapy","authors":"Dongxue Yan, Siqi Bao, Zicheng Zhang, Jie Sun, Meng Zhou","doi":"10.1038/s43588-024-00748-8","DOIUrl":"10.1038/s43588-024-00748-8","url":null,"abstract":"Immune checkpoint inhibitor (ICI) therapies have made considerable advances in cancer immunotherapy, but the complex and diverse spectrum of ICI-induced toxicities poses substantial challenges to treatment outcomes and computational analysis. Here we introduce DySPred, a dynamic graph convolutional network-based deep learning framework, to map and predict the toxicity profiles of ICIs at the population level by leveraging large-scale real-world pharmacovigilance data. DySPred accurately predicts toxicity risks across diverse demographic cohorts and cancer types, demonstrating resilience in small-sample scenarios and revealing toxicity trends over time. Furthermore, DySPred consistently aligns the toxicity-safety profiles of small-molecule antineoplastic agents with their drug-induced transcriptional alterations. Our study provides a versatile methodology for population-level profiling of ICI-induced toxicities, enabling proactive toxicity monitoring and timely tailoring of treatment and intervention strategies in the advancement of cancer immunotherapy. A deep learning framework is proposed with real-world pharmacovigilance data to predict population-scale toxicity profiles of checkpoint inhibitor immunotherapy, enabling proactive toxicity monitoring and timely tailoring of treatment.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 3","pages":"207-220"},"PeriodicalIF":12.0,"publicationDate":"2024-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142883835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deep Bayesian active learning using in-memory computing hardware 使用内存计算硬件的深度贝叶斯主动学习。

IF 12

Nature computational science Pub Date : 2024-12-23 DOI: 10.1038/s43588-024-00744-y

Yudeng Lin, Bin Gao, Jianshi Tang, Qingtian Zhang, He Qian, Huaqiang Wu

{"title":"Deep Bayesian active learning using in-memory computing hardware","authors":"Yudeng Lin, Bin Gao, Jianshi Tang, Qingtian Zhang, He Qian, Huaqiang Wu","doi":"10.1038/s43588-024-00744-y","DOIUrl":"10.1038/s43588-024-00744-y","url":null,"abstract":"Labeling data is a time-consuming, labor-intensive and costly procedure for many artificial intelligence tasks. Deep Bayesian active learning (DBAL) boosts labeling efficiency exponentially, substantially reducing costs. However, DBAL demands high-bandwidth data transfer and probabilistic computing, posing great challenges for conventional deterministic hardware. Here we propose a memristor stochastic gradient Langevin dynamics in situ learning method that uses the stochastic of memristor modulation to learn efficiency, enabling DBAL within the computation-in-memory (CIM) framework. To prove the feasibility and effectiveness of the proposed method, we implemented in-memory DBAL on a memristor-based stochastic CIM system and successfully demonstrated a robot’s skill learning task. The inherent stochastic characteristics of memristors allow a four-layer memristor Bayesian deep neural network to efficiently identify and learn from uncertain samples. Compared with cutting-edge conventional complementary metal-oxide-semiconductor-based hardware implementation, the stochastic CIM system achieves a remarkable 44% boost in speed and could conserve 153 times more energy. This study introduces an in-memory deep Bayesian active learning framework that uses the stochastic properties of memristors for in situ probabilistic computations. This framework can greatly improve the efficiency and speed of artificial intelligence learning tasks, as demonstrated with a robot skill-learning task.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 1","pages":"27-36"},"PeriodicalIF":12.0,"publicationDate":"2024-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11774754/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142883834","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Mapping the gene space at single-cell resolution with gene signal pattern analysis 利用基因信号模式分析绘制单细胞分辨率的基因空间图谱

IF 12

Nature computational science Pub Date : 2024-12-20 DOI: 10.1038/s43588-024-00734-0

Aarthi Venkat, Sam Leone, Scott E. Youlten, Eric Fagerberg, John Attanasio, Nikhil S. Joshi, Michael Perlmutter, Smita Krishnaswamy

{"title":"Mapping the gene space at single-cell resolution with gene signal pattern analysis","authors":"Aarthi Venkat, Sam Leone, Scott E. Youlten, Eric Fagerberg, John Attanasio, Nikhil S. Joshi, Michael Perlmutter, Smita Krishnaswamy","doi":"10.1038/s43588-024-00734-0","DOIUrl":"10.1038/s43588-024-00734-0","url":null,"abstract":"In single-cell sequencing analysis, several computational methods have been developed to map the cellular state space, but little has been done to map or create embeddings of the gene space. Here we formulate the gene embedding problem, design tasks with simulated single-cell data to evaluate representations, and establish ten relevant baselines. We then present a graph signal processing approach, called gene signal pattern analysis (GSPA), that learns rich gene representations from single-cell data using a dictionary of diffusion wavelets on the cell–cell graph. GSPA enables characterization of genes based on their patterning and localization on the cellular manifold. We motivate and demonstrate the efficacy of GSPA as a framework for diverse biological tasks, such as capturing gene co-expression modules, condition-specific enrichment and perturbation-specific gene–gene interactions. Then we showcase the broad utility of gene representations derived from GSPA, including for cell–cell communication (GSPA-LR), spatial transcriptomics (GSPA-multimodal) and patient response (GSPA-Pt) analysis. This work presents a graph signal processing method, gene signal pattern analysis, to embed gene signals from single-cell sequencing data. In diverse experimental set-ups and case studies, GSPA establishes a gene-based framework for single-cell analysis.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"4 12","pages":"955-977"},"PeriodicalIF":12.0,"publicationDate":"2024-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142862464","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A spatiotemporal style transfer algorithm for dynamic visual stimulus generation 一种动态视觉刺激生成的时空风格转移算法。

IF 12

Nature computational science Pub Date : 2024-12-20 DOI: 10.1038/s43588-024-00746-w

Antonino Greco, Markus Siegel

{"title":"A spatiotemporal style transfer algorithm for dynamic visual stimulus generation","authors":"Antonino Greco, Markus Siegel","doi":"10.1038/s43588-024-00746-w","DOIUrl":"10.1038/s43588-024-00746-w","url":null,"abstract":"Understanding how visual information is encoded in biological and artificial systems often requires the generation of appropriate stimuli to test specific hypotheses, but available methods for video generation are scarce. Here we introduce the spatiotemporal style transfer (STST) algorithm, a dynamic visual stimulus generation framework that allows the manipulation and synthesis of video stimuli for vision research. We show how stimuli can be generated that match the low-level spatiotemporal features of their natural counterparts, but lack their high-level semantic features, providing a useful tool to study object recognition. We used these stimuli to probe PredNet, a predictive coding deep network, and found that its next-frame predictions were not disrupted by the omission of high-level information, with human observers also confirming the preservation of low-level features and lack of high-level information in the generated stimuli. We also introduce a procedure for the independent spatiotemporal factorization of dynamic stimuli. Testing such factorized stimuli on humans and deep vision models suggests a spatial bias in how humans and deep vision models encode dynamic visual information. These results showcase potential applications of the STST algorithm as a versatile tool for dynamic stimulus generation in vision science. The spatiotemporal style transfer (STST) algorithm enables video generation by selectively manipulating the spatial and temporal features of natural videos, fostering vision science research in both biological and artificial systems.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":"5 2","pages":"155-169"},"PeriodicalIF":12.0,"publicationDate":"2024-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s43588-024-00746-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142873616","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0