[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-116868-en":3,"doc-seo-116868-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},116868,1099513958762,"Logic","https://ap-avatar.wpscdn.com/avatar/1000023916a998db790?x-image-process=image/resize,m_fixed,w_180,h_180&k=1784791008015729253",8,"Research & Report","Lifelong Machine Learning Potentials - Element-Embracing Atom-Centered Symmetry Functions (eeACSFs)","Machine learning potentials trained on accurate quantum-chemical data can match high accuracy while keeping computational cost low, yet they require system-specific retraining and struggle to encode many chemical elements with common descriptors. This work introduces element-embracing atom-centered symmetry functions (eeACSFs) that merge periodic-table element information with structural features, enabling an element-aware lifelong machine learning potential. Uncertainty quantification supports transferring from a pre-trained fixed model to a continuously adapting lifelong version with assured accuracy, while continual learning methods enable on-the-fly training. A continual resilient (CoRe) optimizer and incremental rehearsal-regularization strategies are proposed for deep neural network training.","arXiv :2303 .059 1 1v2 [ cs .LG] 4 Jun 2023  \nLifelong Machine Learning Potentials  \nMarco Eckhoff∗ and Markus Reiher†  \nETH Zürich, Departement Chemie und Angewandte Biowissenschaften, Vladimir-Prelog-Weg 2, 8093 Zürich, Switzerland.  \n(Dated: May 11, 2023)  \nMachine learning potentials (MLPs) trained on accurate quantum chemical data can retain the high accuracy, while inflicting little computational demands. On the downside, they need to be trained for each individual system. In recent years, a vast number of MLPs has been trained from scratch because learning additional data typically requires to train again on all data to not forget previously acquired knowledge. Additionally, most common structural descriptors of MLPs cannot represent efficiently a large number of different chemical elements. In this work, we tackle these problems by introducing element-embracing atom-centered symmetry functions (eeACSFs) which combine structural properties and element information from the periodic table. These eeACSFs area key for our development of a lifelong machine learning potential (lMLP) . Uncertainty quantification can be exploited to transgress a fixed, pre-trained MLP to arrive at a continuously adapting lMLP, because a predefined level of accuracy can be ensured. To extend the applicability of an lMLP to new systems, we apply continual learning strategies to enable autonomous and on-the-fly training on a continuous stream of new data. For the training of deep neural networks, we propose the continual resilient (CoRe) optimizer and incremental learning strategies relying on rehearsal of data, regularization of parameters, and the architecture of the model.  \nKeywords: Lifelong Machine Learning, Continual Resilient (CoRe) Optimizer, Element-Embracing AtomCentered Symmetry Functions, High-Dimensional Neural Network Potential, Uncertainty Quantification  \n1. INTRODUCTION  \nFor the prediction and understanding of the properties and reactivity of atomistic systems, the knowledge of the potential energy surface is inevitable. The potential energy surface is given by the electronic energy as a function of the nuclear positions and can be obtained from electronic structure methods such as density functional theory (DFT) or approximate wave-function theories [1, 2] . These methods are generally applicable and achieve accurate results for many systems, but even state-of-the-art approaches still lead to high computational demands in extended simulations [3–5] . Instead of explicitly calculating the electronic energy for each nuclear conformation in Born-Oppenheimer approximation, empirical force fields rely on approximate (simple) analytical expressions of the potential energy surface. Hence, they avoid the explicit integration of quantum mechanical equations and thereby enabling efficient atomistic simulations. However, force field expressions are of limited accuracy and they are typically not universal for all chemical systems [6–9] .  \nBy contrast, machine learning potentials (MLPs) [10– 15] can preserve the high accuracy of electronic structure methods, but at low computational cost comparable to that of force fields. MLPs are not based on physical approximations, but rely on very flexible mathematical expressions to obtain an analytical potential energy surface. The parameters of these expressions are trained  \n∗ marco.eckhoff@phys.chem.ethz.ch † [mreiher@ethz.ch](mreiher@ethz.ch)  \non electronic structure reference data including chemical structures and their respective energies and atomic forces. While first-generation MLPs were only applicable to low-dimensional systems [16], the introduction of second-generation MLPs in form of high-dimensional neural network potentials (HDNNPs) [17–19] has facilitated atomistic simulations of systems with tens of thousands of atoms on nanosecond timescales with an accuracy similar to that of first-principles methods. In recent years, various other types of MLPs have been proposed such as neural ","cbCaiuokmRFCWGf4","https://ap.wps.com/l/cbCaiuokmRFCWGf4","pdf",6992829,1,20,"English","en",105,"# Introduction\n## Background: potential energy surfaces and ML potentials\n## Limitations: retraining, element coverage, and extrapolation\n## Active learning and the need for continual learning\n# Core contributions and approach","[{\"question\":\"Why do traditional machine learning potentials require retraining for new systems?\",\"answer\":\"Because their training is typically tied to predefined reference data; incorporating additional data often requires retraining so the model does not forget previously learned knowledge.\"},{\"question\":\"What are element-embracing atom-centered symmetry functions (eeACSFs)?\",\"answer\":\"They combine structural properties with explicit element information from the periodic table, improving efficient representation across many chemical elements for MLPs.\"},{\"question\":\"How is a lifelong machine learning potential made continuously adapting?\",\"answer\":\"Uncertainty quantification enables transferring from a fixed pre-trained MLP to a continuously adapting lifelong version, while continual learning strategies support autonomous on-the-fly training on new incoming data.\"}]","Lifelong Machine Learning Potentials - Element-Embracing Atom-Centered Symmetry Functions (eeACSFs) | PDF",1785672149,50,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"lifelong-machine-learning-potentials-element-embracing-atom-centered-symmetry-functions-eeacsfs","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/lifelong-machine-learning-potentials-element-embracing-atom-centered-symmetry-functions-eeacsfs/116868/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-05","2026-08-02",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"Why do traditional machine learning potentials require retraining for new systems?","Question",{"text":76,"@type":77},"Because their training is typically tied to predefined reference data; incorporating additional data often requires retraining so the model does not forget previously learned knowledge.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What are element-embracing atom-centered symmetry functions (eeACSFs)?",{"text":81,"@type":77},"They combine structural properties with explicit element information from the periodic table, improving efficient representation across many chemical elements for MLPs.",{"name":83,"@type":74,"acceptedAnswer":84},"How is a lifelong machine learning potential made continuously adapting?",{"text":85,"@type":77},"Uncertainty quantification enables transferring from a fixed pre-trained MLP to a continuously adapting lifelong version, while continual learning strategies support autonomous on-the-fly training on new incoming data.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,115,120,123,127,130,134],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":29,"slug":114},6,"Technology","technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":21,"slug":126},9,"Religion & Spirituality","religion-spirituality",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":21,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":107,"slug":137},19,"General","general"]