[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83478-en":3,"doc-seo-83478-105":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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83478,687197100911,"Himbo","https://ap-avatar.wpscdn.com/avatar/a000239b6f1da00475?x-image-process=image/resize,m_fixed,w_180,h_180&k=1782698725881665579",8,"Research & Report","SAOT: Self-Supervised Continual Graph Learning with Structure-Aware Optimal Transport","Self-supervised Continual Graph Learning (CGL) learns from a graph stream across sequential tasks without label supervision. Existing methods mainly enforce instance-level consistency to keep node (or node-pair) embeddings stable, but optimizing nodes independently cannot preserve global relational structure, leading to progressively distorted inter-node correspondences during continual learning. SAOT introduces a Structure-Aware Optimal Transport framework to explicitly capture global node correspondences and preserve relational structure, together with cross-task knowledge distillation. Experiments on four CGL benchmarks show clear gains over self-supervised baselines, up to 5% on CoraFull-CL and over 15% on ProductsCL in Class-IL.","SAOT: Self-Supervised Continual Graph Learning with Structure-Aware  \nOptimal Transport  \nYuting Zhang 1 Yanbei Liu 2 Zhitao Xiao 2 Lei Geng 2 Yanwei Pang 3 Xiao Wang 4  \narXiv :2607 .00377v 1 [ cs .LG] 1 Jul 2026  \nAbstract  \nSelf-supervised Continual Graph Learning (CGL) aims to successively learn from a graph sequence with different tasks without label supervision—a paradigm that has attracted widespread attention. Most existing self-supervised CGL methods rely on instance-level consistency objectives that enforce stability of individual node (or node-pair) embeddings. Due to optimizing nodes in isolation, these methods fail to maintain global relational structure, causing inter-node correspondences to progressively distort under continual learning. To this end, we propose a novel Structure-Aware Optimal Transport (SAOT) framework that explicitly captures and preserves relational structure within graph representations across sequential tasks. Specifically, SAOT leverages optimal transport theory to capture global inter-node correspondences, thereby facilitating and enhancing graph representation learning. Simultaneously, SAOT incorporates a cross-task knowledge distillation mechanism to preserve the previous structural knowledge. Extensive experiments on four CGL benchmark datasets demonstrate that SAOT outperforms existing self-supervised baselines. In particular, SAOT achieves significant performance gains, improving average accuracy by up to 5% on CoraFull-CL and over 15% on ProductsCL compared with state-of-the-art methods in the Class-IL setting.  \n1 School of Electronics and Information Engineering, Tiangong University, Tianjin, China 2 School of Life Sciences, Tiangong  \nUniversity, Tianjin, China 3 School of Electrical and Infomation Engineering, Tianjin University, Tianjin, China 4 School of Software, Beihang University, Beijing, China. Correspondence to: Yanbei Liu \u003C[liuyanbei@tiangong.edu](liuyanbei@tiangong.edu) >.  \nProceedings of the 43 rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026 . Copyright 2026 by the author(s) .  \n1. Introduction  \nGraph data are ubiquitous in real-world applications, modeling complex systems with rich relational dependencies, such as citation networks, e-commerce systems, and biochemical molecules (Hamilton et al., 2017 ; Wu et al., 2021) . Despite the remarkable progress of graph representation learning, most existing methods are designed for static settings, where the data distribution is assumed to be stationary (Kipf & Welling, 2017 ; Velikovi et al., 2018) . In contrast, real-world graph data are continuously generated, with new nodes, edges, or tasks appearing over time (Wanget al., 2020) . For instance, papers on new research topics continuously enter citation networks, and novel molecular properties are progressively encountered in drug discovery tasks (Liu et al., 2021 ; Zhang et al., 2022a) . To cope with such evolving data, graph models are required to incrementally acquire new knowledge while maintaining performance on previously learned tasks, a learning paradigm known as Continual Graph Learning (CGL) (Zhou & Cao, 2021 ; Zhang et al., 2022a) . However, simply training models sequentially on incoming data is prone to catastrophic forgetting (McCloskey & Cohen, 1989 ; Goodfellow et al., 2014), while retraining a model on all accumulated data is computationally expensive and often infeasible when historical data are unavailable.  \nTo mitigate catastrophic forgetting, existing CGL approaches generally are divided into three categories. Parameter isolation methods (Zhang et al., 2023 ; Cai et al., 2022) allocate task-specific parameters or modules to avoid interference between tasks. Regularization-based approaches (Kirkpatrick et al., 2017 ; Liu et al., 2021) constrain parameter updates by penalizing changes to important parameters or outputs learned from previous tasks. Replay-based methods (Zhou & Cao, 2021 ; Zhang et al., 2022b) explicit","cbCaisds6dpeGj2d","https://ap.wps.com/l/cbCaisds6dpeGj2d","pdf",5456955,3,1,13,"English","en",105,"# Abstract\n# Introduction\n## Motivation for Continual Graph Learning\n## Limitations of Existing CGL Approaches\n## Need for Self-Supervised CGL and Structural Preservation\n# Related Work","[{\"question\":\"What is the main goal of self-supervised Continual Graph Learning in this work?\",\"answer\":\"To learn sequentially from a streaming graph with different tasks while avoiding label supervision, improving knowledge retention across tasks.\"},{\"question\":\"Why do instance-level consistency objectives fail in continual graph learning?\",\"answer\":\"They stabilize individual node (or node-pair) embeddings but do not explicitly maintain global relational structure, causing inter-node correspondences to distort over time (structural drift).\"},{\"question\":\"How does SAOT address structural drift during continual learning?\",\"answer\":\"SAOT uses structure-aware optimal transport to capture and preserve global inter-node correspondences and adds cross-task knowledge distillation to retain previous structural knowledge.\"}]",1784188295,33,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"saot-self-supervised-continual-graph-learning-with-structure-aware-optimal-transport","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"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":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/saot-self-supervised-continual-graph-learning-with-structure-aware-optimal-transport/83478/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",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 is the main goal of self-supervised Continual Graph Learning in this work?","Question",{"text":75,"@type":76},"To learn sequentially from a streaming graph with different tasks while avoiding label supervision, improving knowledge retention across tasks.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why do instance-level consistency objectives fail in continual graph learning?",{"text":80,"@type":76},"They stabilize individual node (or node-pair) embeddings but do not explicitly maintain global relational structure, causing inter-node correspondences to distort over time (structural drift).",{"name":82,"@type":73,"acceptedAnswer":83},"How does SAOT address structural drift during continual learning?",{"text":84,"@type":76},"SAOT uses structure-aware optimal transport to capture and preserve global inter-node correspondences and adds cross-task knowledge distillation to retain previous structural knowledge.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"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":52,"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"]