[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86350-en":3,"doc-seo-86350-105":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},86350,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","Asynchronous Perception Machine for Test Time Training","Asynchronous Perception Machine (APM) is proposed as a computationally efficient architecture for test-time training (TTT). APM processes image patches one at a time in arbitrary order while preserving semantic-awareness in the network. It recognizes out-of-distribution images without dataset-specific pre-training, augmentation, or auxiliary pretext tasks, and performs TTT by distilling the test sample representation only once. APM learns from this single representation and also supports scalable semantic clusterings and empirical validation of GLOM’s field-based percept insight.","arXiv :2410 .20535v5 [ cs .CV] 10 Jul 2026  \nAsynchronous Perception Machine for Test Time Training  \nRajat Modi∗ Yogesh Singh Rawat  \nCentre for Research in Computer Vision  \nUniversity of Central Florida  \nOrlando, FL 32765  \n[rajatmodi62@gmail.com](rajatmodi62@gmail.com) ,[yogesh@crcv.ucf.edu](yogesh@crcv.ucf.edu)[ ](yogesh@crcv.ucf.edu)[https://rajatmodi62.github.io/apm_project_page](https://rajatmodi62.github.io/apm_project_page)  \nAbstract  \nIn this work, we propose Asynchronous Perception Machine (APM), a computationally-efficient architecture for test-time-training (TTT) . APM can process patches of an image one at a time in any order asymmetrically, and still encode semantic-awareness in the net. We demonstrate APM’s ability to recognize out-of-distribution images without dataset-specific pre-training, augmentation or any-pretext task. APM offers competitive performance over existing TTT approaches. To perform TTT, APM just distills test sample’s representation once.  \nAPM possesses a unique property: it can learn using just this single representation and starts predicting semantically-aware features.  \nAPM demostrates potential applications beyond test-time-training: APM can scale up to a dataset of 2D images and yield semantic-clusterings in a single forward pass. APM also provides first empirical evidence towards validating GLOM’sinsight, i.e. if input percept is a field. Therefore, APM helps us converge towards an implementation which can do both interpolation and perception on a sharedconnectionist hardware.  \n1 Introduction  \nIn these past centuries, computing-machines have become a lot faster [92] . This made it possible to train higher-parameterized neural nets and led to interesting emergent abilities [80] . As was predicted by Turing himself, and as were his suspicions of Lady Lovelace’s arguments against learning machines[90]: neural-nets can now finally learn without human-feedback [6], paint pictures [76] and even compose a sonnet [92] . Even with such impressive-progress, a key question still remains: how can these nets recognize images whose distribution is far different from the ones which were used during training? For e.g., consider a self-driving car trying to stop when it encounters a pedestrian crossing a road. Such practical scenarios require ‘instantaneous-decisions’ for ensuring human-safety in autonomous-systems [64] .  \nTest-time-training (TTT) [85] is one of the promising techniques for handling such distribution shifts: a neural net adapts to a test sample ‘on the fly’. Since the label of the sample is not known, the net performs some auxiliary pre-text task like data augmentation [15], rotation [15, 86] or prompt tuning [84] on it. After several such iterations, the net recognizes the test sample. The key idea is that the net is allowed to dynamically adjust its decision boundary even after it has been trained, thereby bringing it much closer to how humans keep learning ‘continuously’ throughout their lifespans [86] .  \nDespite the success of existing TTT approaches, several limitations need to be addressed [84, 15, 86]:  \n1) The Information Bottleneck Problem [64]: Multiple TTT iterations requires feed-forward through many hidden layers multiple times, making it computationally expensive. 2) Reliance on a surrogate pre-text task: the optimal data-augmentation pipeline or the best pretext task is not known beforehand, worsening the issue even further in an online setting. 3) Furthermore, TTT leverages  \n∗ [Correspondence to](Correspondence to rajatmodi62@gmail.com)[ rajatmodi62@gmail.com](Correspondence to rajatmodi62@gmail.com).  \n38th Conference on Neural Information Processing Systems (NeurIPS 2024) .  \nFigure 1: (i) Asynchronous Perception Machine (APM): An image I passes through a column module and routes to a trigger column Ti. Ti then unfolds and generates h × w location-specific queries. These queries are i.i.d and can be parallelized across cores of a gpu [48] . Each query Ti is p","cbCaiggDxBPhK3L5","https://ap.wps.com/l/cbCaiggDxBPhK3L5","pdf",5689133,2,1,29,"English","en",105,"# Abstract\n# Introduction\n## Motivation: distribution shift and real-time decision needs\n## Test-time training (TTT) and its limitations\n## Proposed approach: Asynchronous Perception Machine (APM)\n## Contributions","[{\"question\":\"What problem does APM address in test-time training?\",\"answer\":\"APM targets key TTT limitations, especially the information bottleneck caused by repeated forward passes and the reliance on surrogate pretext tasks like augmentation or rotation. It aims to adapt efficiently under distribution shifts.\"},{\"question\":\"How does APM perform test-time training with minimal computation?\",\"answer\":\"APM computes the test sample representation only once. Subsequent TTT iterations overfit using that single representation and do not require augmentation or any-pretext tasks.\"},{\"question\":\"What capability does APM demonstrate beyond standard test-time training?\",\"answer\":\"APM can scale to datasets of 2D images to produce semantic clusterings in a single forward pass. It also provides first empirical evidence supporting GLOM’s insight that a percept behaves like a field.\"}]",1784210736,73,{"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},"asynchronous-perception-machine-for-test-time-training","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/asynchronous-perception-machine-for-test-time-training/86350/",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","2026-07-24","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 problem does APM address in test-time training?","Question",{"text":75,"@type":76},"APM targets key TTT limitations, especially the information bottleneck caused by repeated forward passes and the reliance on surrogate pretext tasks like augmentation or rotation. It aims to adapt efficiently under distribution shifts.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does APM perform test-time training with minimal computation?",{"text":80,"@type":76},"APM computes the test sample representation only once. Subsequent TTT iterations overfit using that single representation and do not require augmentation or any-pretext tasks.",{"name":82,"@type":73,"acceptedAnswer":83},"What capability does APM demonstrate beyond standard test-time training?",{"text":84,"@type":76},"APM can scale to datasets of 2D images to produce semantic clusterings in a single forward pass. It also provides first empirical evidence supporting GLOM’s insight that a percept behaves like a field.","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":20,"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"]