[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122598-en":3,"doc-seo-122598-105":29,"detail-sidebar-cat-0-en-105":86},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":11,"language":21,"language_code":22,"site_id":23,"html_lang":22,"table_of_contents":24,"faqs":25,"seo_title":26,"seo_description":14,"update_tm":27,"read_time":28},122598,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","Continual Conscious Active Fine-Tuning to Robustify Online Machine Learning Models Against Data Distribution Shifts - Research paper overview","Unlike offline learning, online machine learning can handle data distribution shifts during test time, yet current methods remain limited by cost or reliability. This work proposes a continual conscious active fine-tuning augmentation built on test-time adaptation to respond reliably and cost-effectively to drastic distribution changes. The method adds a continual mechanism for ongoing shifts, a conscious scheduling that fine-tunes only when recent shifts are detected, and an active human-machine collaboration step for relabeling. Empirical results show a twofold improvement over traditional test-time adaptation.","Continual Conscious Active Fine-Tuning to Robustify Online Machine Learning Models Against Data Distribution Shifts  \narXiv :2211 .01315v1 [ cs .LG] 2 Nov 2022  \nShawqi Al-Maliki􀀃 , Faissal El Bouanani y, Senior Member, IEEE, Mohamed Abdallah 􀀃 , Senior Member, IEEE, Junaid Qadirz, Senior Member, IEEE, Ala Al-Fuqaha􀀃 , Senior Member, IEEE  \n􀀃 Information and Computing Technology (ICT) Division, College of Science and Engineering, Hamad Bin  \nKhalifa University, Doha 34110, Qatar  \ny College of Engineering, Mohammed V University in Rabat, Morocco z Department of Computer Science and Engineering, College of Engineering, Qatar University, Doha, Qatar  \nAbstract—Unlike their offline traditional counterpart, online machine learning models are capable of handling data distribution shifts while serving at the test time. However, they have limitations in addressing this phenomenon. They are either expensive or unreliable. We propose augmenting an online learning approach called test-time adaptation with a continual conscious active fine-tuning layer to develop an enhanced variation that can handle drastic data distribution shifts reliably and costeffectively. The proposed augmentation incorporates the following aspects: a continual aspect to confront the ever-ending data distribution shifts, a conscious aspect to imply that fine-tuning is a distribution-shift-aware process that occurs at the appropriate time to address the recently detected data distribution shifts, and an active aspect to indicate employing human-machine collaboration for the relabeling to be cost-effective and practical for diverse applications. Our empirical results show that the enhanced testtime adaptation variation outperforms the traditional variation by a factor of two.  \nIndex Terms—Online Machine Learning, data distribution shifts, conscious fine-tuning, robust ML models, next-generation AI, socio-technical human-machine collaboration.  \nI. INTRODUCTION  \nThe tremendous success of Machine Learning (ML) models is conditioned on the assumption that training and inference time data come from the same distribution. Breaking this assumption, which is the case in many real-world applications, renders ML models susceptible to data distribution shifts. The data distribution shifts occur whenever ML model gets trained on data with a certain distribution and applied to data with a different distribution [1] . It may exist naturally as a result of dynamic changes in the process (common in many real-world ML applications in the wild) or may be induced adversarially for the explicit purpose of compromising the model's performance. Regardless of the source of data distribution shift, the performance of ML models is severely compromised by data distribution shifts. In this work, we address the natural  \ndata distribution shifts on the image classification task as a representative proxy for handling different sources of data distribution shifts. Addressing the concern of data distribution shift, effectively and efficiently, is a must to maintain ML models robust and well-performing.  \nTraditional models (current AI) are the most ubiquitous form of ML models. They have offline settings, in which ML models get trained using offline data and do not get updated at test time. In the offline setting, data distribution shifts are usually addressed by blind retraining, which occurs at different time intervals using a large amount of data. However, this is an inefficient and unreliable retraining process as it is costly and has a low potential to be performed at the right time. Thus, offline-learned ML models are inappropriate for handling distribution shifts effectively. This drawback restricts their use in mission-critical applications. Moreover, we live inan era of ever-changing data and disruptive technologies: IoT, smart cities, cyber-physical systems, metaverse, and industry 4.0 are the most disruptive and promising technologies [2] [3] . That makes addressing offline-learned ML models' d","cbCaivQLkMdVZJSu","https://ap.wps.com/l/cbCaivQLkMdVZJSu","pdf",734476,1,"English","en",105,"# Introduction\n## Current Online Learning Approaches and Limitations\n## Online Supervised Learning - cost and label availability\n## Test-time Adaptation - limits under drastic shifts","[{\"question\":\"What is the proposed solution?\",\"answer\":\"It combines continual handling of ongoing shifts, shift-aware timing for fine-tuning, and active human-machine collaboration for relabeling.\"},{\"question\":\"How does the proposed method relate to existing online learning paradigms?\",\"answer\":\"The approach aims to better support real-world applications by improving robustness against distribution changes.\"}]","Continual Conscious Active Fine-Tuning to Robustify Online Machine Learning Models Against Data Distribution Shifts - Research paper overview | PDF",1785811662,20,{"code":4,"msg":30,"data":31},"ok",{"site_id":23,"language":22,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":81,"head_meta":83,"extra_data":85,"updated_unix":27},"continual-conscious-active-fine-tuning-to-robustify-online-machine-learning-models-against-data-distribution-shifts-research-paper-overview","",{"@graph":35,"@context":80},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/continual-conscious-active-fine-tuning-to-robustify-online-machine-learning-models-against-data-distribution-shifts-research-paper-overview/122598/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":22,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-08-04",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76],{"name":71,"@type":72,"acceptedAnswer":73},"What is the proposed solution?","Question",{"text":74,"@type":75},"It combines continual handling of ongoing shifts, shift-aware timing for fine-tuning, and active human-machine collaboration for relabeling.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the proposed method relate to existing online learning paradigms?",{"text":79,"@type":75},"The approach aims to better support real-world applications by improving robustness against distribution changes.","https://schema.org",{"og:url":51,"og:type":82,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":84,"canonical":51},"index,follow",{"doc_id":7,"site_id":23},{"code":4,"msg":5,"data":87},[88,92,96,100,105,110,115,118,122,125,129],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":89,"show_sort_weight":90,"slug":91},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Exam",70,"exam",{"id":101,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},5,"Comic",60,"comic",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},6,"Technology",50,"technology",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":116,"slug":117},30,"research-report",{"id":119,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":28,"slug":121},9,"Religion & Spirituality","religion-spirituality",{"id":28,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":28,"slug":124},"World Cup","world-cup",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":126,"slug":128},10,"Lifestyle","lifestyle",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":101,"slug":132},19,"General","general"]