[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-123072-en":3,"doc-seo-123072-105":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":4,"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},123072,8796095461610,"Oliver","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","A Pattern Language for Machine Learning Tasks","Idealised as universal approximators, neural learners can be interpreted as variable functions that become concrete functions after training. Objective functions are treated as constraints on learner behaviour, and perfectly optimised equivalences are extracted as “tasks”. A formal graphical language is developed to separate task intent from implementation, enable model-agnostic reasoning and design, and unify machine-learning approaches across domains. A proof-of-concept “manipulator” task converts classifiers into generative models supporting style transfer and interpretable latent editing without bespoke architectures.","arXiv :2407 .02424v 1 [ cs .LG] 2 Jul 2024  \nA Pattern Language for Machine Learning Tasks  \nBenjamin Rodatz∗‡† Ian Fan∗† Tuomas Laakkonen†  \nNeil John Ortega† Thomas Hoffman† Vincent Wang-Maścianica∗†‡\\#  \n†Compositional Intelligence, Quantinuum  \n17 Beaumont St., Oxford OX1 2NA, UK  \n[first.last@quantinuum.com](first.last@quantinuum.com)  \n‡Department of Computer Science, University of Oxford  \n7 Parks Rd, Oxford OX1 3QG, UK  \nAbstract  \nIdealised as universal approximators, learners such as neural networks can be viewed as“variable functions” that may become one of a range of concrete functions after training. In the same way that equations constrain the possible values of variables in algebra, we may view objective functions as constraints on the behaviour of learners. We extract the equivalences perfectly optimised objective functions impose, calling them “tasks”. For these tasks, we develop a formal graphical language that allows us to: (1) separate the core tasks of a behaviour from its implementation details; (2) reason about and design behaviours model-agnostically; and (3) simply describe and unify approaches in machine learning across domains.  \nAs proof-of-concept, we design a novel task that enables converting classifiers into generative models we call “manipulators”, which we implement by directly translating task specifications into code. The resulting models exhibit capabilities such as style transfer and interpretable latent-space editing, without the need for custom architectures, adversarial training or random sampling. We formally relate the behaviour of manipulators to GANs, and empirically demonstrate their competitive performance with VAEs. We report on experiments across vision and language domains aiming to characterise manipulators as approximate Bayesian inversions of discriminative classifiers.  \n1 Introduction  \nThe primary instrument for controlling the training of machine learning (ML) models is the objective function, which we can view as a map from possible parameters of the model to a real number we call the loss. Designing a good objective function incorporates many empirical choices, such as choice of architecture, measure of statistical divergence, and training data (Ciampiconi et al. 2023; Richardson 2022; Terven et al. 2023), that are often chosen by heuristics, and rationalised post hoc. Though many of these choices are required to make training tractable, they are not relevant to understanding the final behaviour of the trained model. Instead, we propose that an essential to understanding model behaviour is the specification of the behavioural equivalences achieved if the objective function is perfectly optimised.  \n*Equal contribution \\# Corresponding author ([vincent.wang@quantinuum.com](vincent.wang@quantinuum.com))  \nAn objective function can be broken into three parts. Let Θ be the parameter-space of a model 􀁦 . Then a generic supervised classification task — for example, labelling images — amounts to minimising (for 􀀒 ∈ Θ) the following objective function:  \n􀁅 [􀁄 (􀁦 􀀒 (􀁩􀁭), 􀁬􀁡􀁢 )]  \n(􀁩􀁭,􀁬􀁡􀁢)∼􀁘  \nFirstly, we take the expected value over some data distribution of paired images 􀁩􀁭 with their labels 􀁬􀁡􀁢 . Secondly, we choose a measure of statistical divergence 􀁄, such as cross-entropy or log-likelihood. Finally, we have the two sides that, under perfect conditions, should be equal: the output of the classifier 􀁦 􀀒 given an image 􀁩􀁭 and the matching label 􀁬􀁡􀁢 . In this work, we focus on the final aspect: the two things we want to be equal. We call this equational constraint a task to distinguish it from the objective function. Further, we abstract away implementation details, such as architecture and training, by idealising models to be universal function approximators that can, in principle, perfectly optimise objective functions. Thus each task can be viewed purely as an equational constraint on the behaviour of the learners, comparable to equational constraints on the possible value","cbCail9KIAlXcDmq","https://ap.wps.com/l/cbCail9KIAlXcDmq","pdf",1909044,1,26,"English","en",105,"# Introduction\n# Tasks and patterns\n## Mathematical preliminaries","[{\"question\":\"What does the paper mean by treating objective functions as constraints on learners?\",\"answer\":\"Objective functions constrain which outputs a trained model can produce. Under perfect optimisation, the behaviour equivalences enforced by an objective function can be extracted as tasks.\"},{\"question\":\"What is the purpose of the proposed formal graphical language?\",\"answer\":\"It provides a way to represent and reason about tasks separately from implementation details, enabling model-agnostic behaviour design and unification of approaches across machine-learning domains.\"},{\"question\":\"How do the authors demonstrate the approach with “manipulators”?\",\"answer\":\"They design a task that translates task specifications into code, converting classifiers into generative models. The resulting models achieve effects like style transfer and interpretable latent-space editing without custom architectures, adversarial training, or random sampling.\"}]","A Pattern Language for Machine Learning Tasks | PDF",1785814516,66,{"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},"a-pattern-language-for-machine-learning-tasks","",{"@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/a-pattern-language-for-machine-learning-tasks/123072/",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":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What does the paper mean by treating objective functions as constraints on learners?","Question",{"text":75,"@type":76},"Objective functions constrain which outputs a trained model can produce. Under perfect optimisation, the behaviour equivalences enforced by an objective function can be extracted as tasks.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What is the purpose of the proposed formal graphical language?",{"text":80,"@type":76},"It provides a way to represent and reason about tasks separately from implementation details, enabling model-agnostic behaviour design and unification of approaches across machine-learning domains.",{"name":82,"@type":73,"acceptedAnswer":83},"How do the authors demonstrate the approach with “manipulators”?",{"text":84,"@type":76},"They design a task that translates task specifications into code, converting classifiers into generative models. The resulting models achieve effects like style transfer and interpretable latent-space editing without custom architectures, adversarial training, or random sampling.","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"]