[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-119584-en":3,"doc-seo-119584-105":30,"detail-sidebar-cat-0-en-105":92},{"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":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},119584,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","Deep Generative Models as an Adversarial Attack Strategy for Tabular Machine Learning","Deep generative models are used to produce synthetic data, and they can be repurposed to craft adversarial examples that test the robustness of machine learning systems. Extending adversarial DGMs to tabular ML is difficult because tabular features require correct preprocessing and must satisfy domain constraints. This paper converts four popular tabular DGMs into adversarial DGMs, adding a constraint repair layer to ensure outputs remain realistic while changing model predictions with minimal deviation.","DEEP GENERATIVE MODELS AS AN ADVERSARIAL ATTACK STRATEGY  \nFOR TABULAR MACHINE LEARNING  \narXiv :2409 . 12642v1 [ cs .LG] 19 Sep 2024  \nSALIJONA DYRMISHI 1 , MIHAELA C˘AT˘ALINA STOIAN2 , ELEONORA GIUNCHIGLIA3 , MAXIME CORDY 1  \n1University of Luxembourg, Luxembourg  \n2University of Oxford, United Kingdom  \n3Imperial College London, United Kingdom  \nE-MAIL: [salijona.dyrmishi@uni.lu](salijona.dyrmishi@uni.lu), [mihaela.stoian@st-hildas.ox.ac.uk](mihaela.stoian@st-hildas.ox.ac.uk)  \n[e.giunchiglia@imperial.ac.uk](e.giunchiglia@imperial.ac.uk), [maxime.cordy@uni.lu](maxime.cordy@uni.lu)  \nAbstract:  \nDeep Generative Models (DGMs) have found application in computer vision for generating adversarial examples to test the robustness of machine learning (ML) systems. Extending these adversarial techniques to tabular ML presents unique challenges due to the distinct nature of tabular data and the necessity to preserve domain constraints in adversarial examples. In this paper, we adapt four popular tabularDGMs into adversarial DGMs (AdvDGMs) and evaluate their effectiveness in generating realistic adversarial examples that conform to domain constraints.  \nKeywords:  \nadversarial attacks, deep generative models, tabular ML  \n1. Introduction  \nDeep Generative Models (DGMs) generate synthetic data after learning the probability distribution of their training data. They are most commonly used to augment datasets for better predictive performance of machine learning (ML) models [1], to promote fairness [2], ensure privacy [3] etc. Several works in computer vision have repurposed DGMs as a tool to generate adversarial examples for ML models. Such adversarial examples pose a significant security threat by minimally altering original inputs and forcing the models to wrongfully change their predictions. In this scenario, adversarial DGMs (AdvDGMs) take as an input an original example and output an adversarial one while trying to minimize the adversarial and perturbation loss, in addition to their original loss functions. AdvDGMs promise shorter generation times compared to today’s popular iterative adversarial attacks, which is beneficial and important for adversarial hardening [4] .  \nBeyond computer vision, extending AdvDGMs to test and improve the robustness of tabular ML models against adversarial examples is challenging due to the unique characteristics of tabular data. These models must account for diverse feature types and preprocessing while ensuring the generated adversarial examples adhere to domain-specific constraints. For instance, in a credit scoring system, the “average transaction amount” must not exceed the “maximum transaction amount.”Violating such constraints results in unrealistic examples that do not map to real-world transaction history. Current tabular DGMs often fail in this regard, producing up to 100% unrealistic examples [5] . Few efforts have been made to adapt DGMs into AdvDGMs for tabular data [6, 7, 8, 9], however, these attempts often focus on a single use case, use generic models not tailored for tabular data, and handle the realism of their outputs by modifying only independent features, thus limiting the adversarial example search space.  \nIn this paper, we convert four popular tabular DGMs into AdvDGMs and evaluate their potential to generate successful adversarial examples that fulfill three objectives: satisfy constraints, change model prediction, and maintain minimal distance from the original input. To boost the performance of tabular AdvDGMs, we extend them with a constraint repair layer [5], ensuring that the outputs always satisfy domain constraints. Adding the constraint repair layer should not significantly impact the efficiency of AdvCDGMs compared to iterative adversarial techniques. Hence, we investigate the impact of CL on runtime. Finally, we compare our AdvDGMs’ performance with three attacks from literature optimized for domain constraints. The source code, the data and the models are publicly","cbCait83h8zy8J6Q","https://ap.wps.com/l/cbCait83h8zy8J6Q","pdf",571892,1,6,"English","en",105,"# Introduction\n# Related Work\n## DGMs for Tabular Data\n## DGMs as an attack strategy","[{\"question\":\"Why is extending adversarial deep generative models to tabular machine learning challenging?\",\"answer\":\"Tabular data contain diverse feature types and preprocessing steps, and adversarial examples must respect domain-specific constraints. Violations lead to unrealistic samples that do not reflect real-world data.\"},{\"question\":\"What does the paper propose to build adversarial DGMs for tabular data?\",\"answer\":\"The work adapts four popular tabular DGMs into adversarial DGMs that output adversarial examples intended to satisfy constraints, change the model prediction, and stay close to the original input.\"},{\"question\":\"How does the constraint repair layer improve adversarial example realism?\",\"answer\":\"A constraint repair layer enforces domain constraints on generated outputs, preventing unrealistic examples and aiming to preserve efficient performance relative to iterative adversarial techniques.\"}]","Deep Generative Models as an Adversarial Attack Strategy for Tabular Machine Learning | PDF",1785725121,15,{"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":87,"head_meta":89,"extra_data":91,"updated_unix":28},"deep-generative-models-as-an-adversarial-attack-strategy-for-tabular-machine-learning","",{"@graph":36,"@context":86},[37,54,69],{"@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/deep-generative-models-as-an-adversarial-attack-strategy-for-tabular-machine-learning/119584/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-04","2026-08-03",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"Why is extending adversarial deep generative models to tabular machine learning challenging?","Question",{"text":76,"@type":77},"Tabular data contain diverse feature types and preprocessing steps, and adversarial examples must respect domain-specific constraints. Violations lead to unrealistic samples that do not reflect real-world data.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What does the paper propose to build adversarial DGMs for tabular data?",{"text":81,"@type":77},"The work adapts four popular tabular DGMs into adversarial DGMs that output adversarial examples intended to satisfy constraints, change the model prediction, and stay close to the original input.",{"name":83,"@type":74,"acceptedAnswer":84},"How does the constraint repair layer improve adversarial example realism?",{"text":85,"@type":77},"A constraint repair layer enforces domain constraints on generated outputs, preventing unrealistic examples and aiming to preserve efficient performance relative to iterative adversarial techniques.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},"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":107,"slug":138},19,"General","general"]