[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-327901-105":59,"doc-detail-327901-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","hierarchical-multi-to-single-modal-knowledge-distillation-for-disruption-prediction-in-east","Hierarchical Multi-to-Single-Modal Knowledge Distillation for Disruption Prediction in EAST","","Plasma disruption poses a critical threat to tokamak safety, and current data-driven predictors primarily use time-series diagnostic signals, while visible images add complementary spatial cues but increase inference cost. A hierarchical multi-to-single-modal knowledge distillation framework is proposed for disruption prediction on a synchronized EAST multimodal dataset. A Transformer-based multimodal teacher is trained with prototype-guided spatiotemporal hypergraphs, then distilled to a time-series-only student via graph, representation, and decision-level distillation. On the 640-discharge EAST dataset, the teacher reaches TPR 100% and AUC 99%, while the distilled student attains TPR 91.66% and AUC 97.88% with much lower inference time and reduced FLOPs/parameters, demonstrating robustness and scalability for efficient EAST disruption prediction.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/hierarchical-multi-to-single-modal-knowledge-distillation-for-disruption-prediction-in-east/327901/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/hierarchical-multi-to-single-modal-knowledge-distillation-for-disruption-prediction-in-east/327901.png","ImageObject",300,407,{"name":92,"@type":93},"Levi","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-03","2026-09-21",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",13,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does the paper address in EAST disruption prediction?","Question",{"text":113,"@type":114},"It addresses the high inference cost caused by using visible images, aiming to retain multimodal discriminative benefits while performing inference using only time-series signals on EAST.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How is the hierarchical knowledge distillation framework structured?",{"text":118,"@type":114},"A multimodal teacher is trained using visible images and time-series signals, then a time-series-only student receives transferred knowledge through graph-structure-level, representation-level, and decision-level distillation.",{"name":120,"@type":111,"acceptedAnswer":121},"What performance and efficiency results are reported on the EAST dataset?",{"text":122,"@type":114},"On the 640-discharge EAST dataset, the teacher achieves TPR 100% and AUC 99%, while the distilled student achieves TPR 91.66% and AUC 97.88% with faster inference, reduced FLOPs by about 68.90%, and fewer parameters by about 47.85%.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},327901,1790465501,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":145,"read_time":146,"preview_page_count":4},7971461740909,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","arXiv :2607 .0424 1v 1 [ cs .CV] 5 Jul 2026  \nHierarchical Multi-to-Single-Modal Knowledge Distillation for Disruption Prediction in EAST  \nQiang Chen 1 , Xiao Wang 1 , Hao Si 1 , Qingquan Yang2* , Meiwen Chen3 , Jianhua Yang2* , Xiaofeng Han2 , Yunhu Jia2 , Ran Chen2 , Liang Wang2 , Jin Tang 1 , Guosheng Xu2  \n1 School of Computer Science and Technology, Anhui University, Hefei 230031, China  \n2 Institute of Plasma Physics, Chinese Academy of Sciences, Hefei 230031, China  \n3 State Key Laboratory of Opto-Electronic Information Acquisition and Protection Technology, Anhui University, Hefei 230039, China  \nE-mail: {yangqq, [yangjh](yangjh}@ipp.ac.cn)[}](yangjh}@ipp.ac.cn)[@ipp.ac.cn](yangjh}@ipp.ac.cn)  \nAbstract. Plasma disruption is a critical threat to tokamak safety. Existing datadriven predictors mainly rely on time-series diagnostic signals, while visible images provide complementary spatial cues including plasma deformation, local brightening, and radiation-structure evolution. Although the image modality improves the model’s discriminative capability, it also substantially increases the computational cost during inference. To address this issue, we propose a hierarchical multito-single-modal knowledge distillation framework for disruption prediction on a synchronized EAST multimodal dataset. During training, visible images and timeseries signals are used to train a multimodal teacher, which learns disruption precursor representations through Transformer-based encoders and a prototype-guided spatiotemporal hypergraph module. During inference, only the time-series student is retained, with multimodal knowledge transferred through graph-structure-level, representation-level, and decision-level distillation. On the 640-discharge EAST dataset, using the validation-selected thresholds with a 10 ms valid-warning constraint, the multimodal teacher achieves strong performance, with a TPR of 100.00%, an FPRof 2 .73%, an F1 score of 96 .00%, and an AUC of 99 .00% . The distilled unimodal student, which relies on time-series inputs during inference, still achieves a TPR of 91.66%, an FPR of 2.73%, an F1 score of 91.66%, and an AUC of 97.88%, while requiring only 3.75 ms for inference, 9.75G FLOPs, and 30.26M parameters. Compared with the multimodal teacher, the distilled student achieves a 2 .16 × inference speedup and reduces the FLOPs and parameter count by approximately 68.90% and 47.85%, respectively. Experiments on the independent generalization dataset and the largerscale dataset demonstrate the robustness and scalability of the proposed framework, while the similarity analysis confirms that the distilled student effectively inherits the teacher’s knowledge in latent representation, decision boundary, and structural preference. These results demonstrate that the proposed framework can preserve the discriminative advantages of multimodal learning while substantially reducing inference cost, and provide an effective route for efficient disruption prediction in EAST. The source code of this paper will be released on [https://github.com/](https://github.com/)[ ](https://github.com/)[Event-AHU/OpenFusion](Event-AHU/OpenFusion.)[.](Event-AHU/OpenFusion.)  \nHierarchical Multi-to-Single-Modal Knowledge Distillation for Disruption Prediction in EAST2 Keywords: disruption prediction, deep learning, multimodal, EAST tokamak  \n1. Introduction  \nPlasma disruption is one of the major operational challenges in tokamak devices. A disruption is accompanied by thermal quench, current quench, runaway electrons, and strong electromagnetic and thermal loads, all of which may cause substantial damage to the first wall, divertor, and other plasma-facing components [1 , 2] . As future devices move toward larger size, higher plasma current, and higher stored energy, the operational risk associated with unmitigated disruptions will further increase. Therefore, reliable disruption prediction capable of issuing alarms with sufficient lead time remains a","cbCaieYvhD3xP5Qe","https://ap.wps.com/l/cbCaieYvhD3xP5Qe","pdf",13002957,26,"English","# Introduction\n## Plasma disruption risk and need for reliable alarms\n## Existing time-series-based disruption prediction paradigms\n### High-performance single-device prediction\n### Cross-device transfer and few-shot deployment\n### Interpretability and plasma-state representation modeling","[{\"question\":\"What problem does the paper address in EAST disruption prediction?\",\"answer\":\"It addresses the high inference cost caused by using visible images, aiming to retain multimodal discriminative benefits while performing inference using only time-series signals on EAST.\"},{\"question\":\"How is the hierarchical knowledge distillation framework structured?\",\"answer\":\"A multimodal teacher is trained using visible images and time-series signals, then a time-series-only student receives transferred knowledge through graph-structure-level, representation-level, and decision-level distillation.\"},{\"question\":\"What performance and efficiency results are reported on the EAST dataset?\",\"answer\":\"On the 640-discharge EAST dataset, the teacher achieves TPR 100% and AUC 99%, while the distilled student achieves TPR 91.66% and AUC 97.88% with faster inference, reduced FLOPs by about 68.90%, and fewer parameters by about 47.85%.\"}]","Hierarchical Multi-to-Single-Modal Knowledge Distillation for Disruption Prediction in EAST | PDF",1789978298,66]