[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-121958-en":3,"doc-seo-121958-105":30,"detail-sidebar-cat-0-en-105":91},{"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},121958,4398048950312,"Violet","https://ap-avatar.wpscdn.com/avatar/400002538284de19e3c?_k=1778320343897328908",8,"Research & Report","Adaptive Loss Weighting for Machine Learning Interatomic Potentials","Training machine learning interatomic potentials requires optimizing a loss function with three terms: potential energies, atomic forces, and stress. Standard practice uses fixed coefficients for these terms, often chosen through iterative or heuristic procedures that can lead to non-optimal training outcomes. An adaptive loss weighting algorithm is proposed to automatically adjust the weights during training based on dataset characteristics, yielding more balanced predictions across energy, force, and stress while improving overall prediction accuracy.","arXiv :2403 . 18122v1 [physics .comp-ph] 26 Mar 2024  \nAdaptive Loss Weighting for Machine Learning Interatomic Potentials  \nDaniel Ocampoa , Daniela Possob , Reza Namakiana , Wei Gaoa,c,∗  \naJ. Mike Walker ′ 66 Department of Mechanical Engineering, Texas A&M University, College Station, Texas 77843, United  \nStates  \nb Department of Mechanical Engineering, The University of Texas at San Antonio, San Antonio, Texas 78249, United States c Department of Materials Science & Engineering, Texas A&M University, College Station, Texas 77843, United States  \nAbstract  \nTraining machine learning interatomic potentials often requires optimizing a loss function composed of three variables: potential energies, forces, and stress. The contribution of each variable to the total loss is typically weighted using fixed coefficients. Identifying these coefficients usually relies on iterative or heuristic methods, which may yield sub-optimal results. To address this issue, we propose an adaptive loss weighting algorithm that automatically adjusts the loss weights of these variables during the training of potentials, dynamically adapting to the characteristics of the training dataset. The comparative analysis of models trained with fixed and adaptive loss weights demonstrates that the adaptive method not only achieves amore balanced predictions across the three variables but also improves overall prediction accuracy. Keywords: machine learning, interatomic potentials, adaptive learning rate, loss function, neural network  \n1. Introduction  \nMachine learning inter-atomic potentials (ML-IAPs) can be broadly split into two types. The first is descriptor-based ML-IAP, in which the descriptors (or fingerprints) are used to describe the environment of the atoms in a system. Various descriptors have been proposed in the literature, such as Atom-Centered Symmetry Functions (ACSF) [1], Smooth Overlap of Atomic Positions (SOAP), Atomic Cluster Expansion (ACE) [2], and Moment Tensor Potentials [3], among others. A comprehensive review of the descriptors can be found in Musil et al.’s work [4] . Representative descriptor-based ML-IAPs include: Behler and Parrinello Neural Network potential [1, 5], Gaussian approximation potential (GAP) [6], Spectral Neighbor Analysis Potential (SNAP) [7], Moment Tensor Potential (MTP) [3], Performant implementation of the atomic cluster expansion (PACE) [8], and DeePMD [9] among others. The second type of ML-IAP is the end-to-end potential, which operates differently by learning directly from the types and positions of atoms, without the need for predefined descriptors. Representative ones include: Crystal Graph Convolutional Neural Networks (CGCNN) [10, 11], SchNet [12], and MatErials Graph Network (MEGNet) [13] among others. Although the end-to-end ML-IAPs leverage more recent and advanced feature learning AI technology, there is currently no  \n∗ Corresponding author  \nEmail address: [wei.gao@tamu.edu](wei.gao@tamu.edu) (Wei Gao)  \nconclusive evidence to suggest that end-to-end ML-IAPs outperform the descriptor-based ML-IAP in terms of prediction accuracy. The study reported in this paper are performed using ACSF.  \nThe training of most ML-IAPs involves minimizing a loss function, which measures the difference between the predicted outputs of the potential and the actual target value obtained from Ab initio simulations. Typically, an ML-IAP’s loss function comprises three components: potential energy, atomic forces, and stress tensor, each weighted by a prefactor (i.e. loss weight) . Most of current ML-IAPs assign a predefined loss weight to each component, which stays as a constant throughout the training process [3, 5 , 7 , 8] . The approaches like DeepMD modulate these weights linearly during training, though the rationale and effectiveness of this approach are not fully clear [9, 14] . In this paper, our study demonstrates that varying combinations of loss weights significantly impact model performance, raising a key q","cbCaihRdAuID3Rz0","https://ap.wps.com/l/cbCaihRdAuID3Rz0","pdf",731941,1,13,"English","en",105,"# Introduction\n## ML interatomic potential types and loss function components\n# Computation Methods\n## Adaptive loss weighting algorithm implementation","[{\"question\":\"What problem does adaptive loss weighting address in ML interatomic potentials?\",\"answer\":\"It addresses the fact that using fixed loss weights for energy, forces, and stress can be sub-optimal, since the optimal balance depends on the training dataset and material system.\"},{\"question\":\"Which loss components are adjusted during training?\",\"answer\":\"The method adaptively adjusts the weights of the three loss components: potential energies, atomic forces, and the stress tensor.\"},{\"question\":\"How is the adaptive algorithm evaluated against fixed-weight training?\",\"answer\":\"Models trained with adaptive loss weights are compared to models trained with fixed weights, showing better balance across energy, forces, and stress and improved overall prediction accuracy.\"}]","Adaptive Loss Weighting for Machine Learning Interatomic Potentials | PDF",1785808028,33,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":28},"adaptive-loss-weighting-for-machine-learning-interatomic-potentials","",{"@graph":36,"@context":85},[37,54,68],{"@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/adaptive-loss-weighting-for-machine-learning-interatomic-potentials/121958/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-04",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does adaptive loss weighting address in ML interatomic potentials?","Question",{"text":75,"@type":76},"It addresses the fact that using fixed loss weights for energy, forces, and stress can be sub-optimal, since the optimal balance depends on the training dataset and material system.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Which loss components are adjusted during training?",{"text":80,"@type":76},"The method adaptively adjusts the weights of the three loss components: potential energies, atomic forces, and the stress tensor.",{"name":82,"@type":73,"acceptedAnswer":83},"How is the adaptive algorithm evaluated against fixed-weight training?",{"text":84,"@type":76},"Models trained with adaptive loss weights are compared to models trained with fixed weights, showing better balance across energy, forces, and stress and improved overall prediction accuracy.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"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":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]