[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-150705-en":3,"doc-seo-150705-105":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},150705,13056703020460,"Valentina","https://ap-avatar.wpscdn.com/avatar/be000253dac470eee5d?_k=1778207105932848923",8,"Research & Report","Interpreting the Latent Space of Generative Adversarial Networks using Supervised Learning","With rapid progress in Generative Adversarial Networks (GANs), understanding and manipulating GAN latent space has become central to many applications. Prior work often relies on unsupervised learning, which makes training difficult and limits interpretability. This work introduces a supervised approach that encodes human prior knowledge via label space, learning a linear mapping from latent variables to attributes. The method uses orthogonality regularization to reduce coefficient correlation, enabling accurate and robust image manipulation results.","Interpreting the Latent Space of Generative Adversarial Networks using Supervised Learning  \nToan Pham Van 1 , Tam Minh Nguyen 1 , Ngoc N. Tran 1 , Hoai Viet Nguyen 1 ,  \nLinh Bao Doan 1 , Huy Quang Dao 1 , Thanh Ta Minh 1 ;2  \n1R&D Lab, Sun* Inc  \n{pham.van.toan, nguyen.minh.tamb, [tran.ngo.quang.ngoc](tran.ngo.quang.ngoc), nguyen.viet.hoai, doan.bao.linh, dao.quang.huyb, [ta.minh.thanh}@sun-asterisk.com](ta.minh.thanh}@sun-asterisk.com)  \n2Le Quy Don Technical University, 236 Hoang Quoc Viet, Bac Tu Liem, Ha Noi  \n[thanhtm@mta.edu.vn](thanhtm@mta.edu.vn)  \narXiv :2102 . 12139v1 [ cs .LG] 24 Feb 2021  \nAbstract—With great progress in the development of Generative Adversarial Networks (GANs), in recent years, the quest for insights in understanding and manipulating the latent space of GAN has gained more and more attention due to its wide range of applications. While most of the researches on this task have focused on unsupervised learning method, which induces difficulties in training and limitation in results, our work approaches another direction, encoding human's prior knowledge to discover more about the hidden space of GAN. With this supervised manner, we produce promising results, demonstrated by accurate manipulation of generated images. Even though our model is more suitable for task-specific problems, we hope that its ease in implementation, preciseness, robustness, and the allowance of richer set of properties (compared to other approaches) for image manipulation can enhance the result of many current applications.  \nIndex Terms—Latent space, Generative Adversarial Networks, Orthogonality Regularization, Supervised Learning  \nI. INTRODUCTION  \nThe task of image generation has introduced many interesting applications in the world of computer vision, including image-to-image translation [1], [2], character drawing generation [3], and more. In recent years, a lot of effort has been done into enhancing Generative Adversarial Networks (GAN) [4] -a model that produced very promising results for the above applications. The main objective of this model is to generate realistic data from a random vector, in much lower dimension, sampled from a prior distribution. Called the “feature vector”, this random vector is believed to have encoded and condensed the important characteristics of the image; and the task of GAN is to generate an image from that information.  \nAs the quality of image samples increases, more attention has also been drawn to latent space interpretation and image manipulation. This task studies how interactions between univariables in the latent space z of GAN results in generated images that we observe, particularly how to sample images with desired properties or manipulate attributes of synthesized images. The task can be usefully applied in Photograph Editing [5], Face Aging [6],. . .  \nOverall, many works on learning the latent space share a main limitation: they do not encode human's prior knowledge into their studies. If one has some prior knowledge about  \nthe domain, they can appropriately assume several factors of variation of the data. Taking human faces as an example, it's common to presume that a successful GAN model's latent space can encode numerous attributes including: faces pose, hair's colors, attractiveness, baldness, smiling... This knowledge can be useful for task-specific training. Many approaches can be applied more generally; however, they can be either very time-consuming and difficult to train (such as InfoGAN [7]) or unable to learn specific factors of variation at all (for instance, vector arithmetic [8] and interpolation [8] between two images can only show that latent space indeed encodes meaningful semantics) .  \nIn this paper, we propose a method of learning the hidden latent space, and as a direct result, meaningfully manipulating GAN model`s generated images. Specifically, since the human prior knowledge is in the form of labels provided with the data, we shall call the domain of t","cbCaikfFRllNXBnZ","https://ap.wps.com/l/cbCaikfFRllNXBnZ","pdf",972431,1,6,"English","en",105,"# Introduction\n## Latent space interpretation and image manipulation\n## Motivation: limitations of unsupervised approaches\n## Proposed supervised mapping and orthogonality regularization","[{\"question\":\"What is the main goal of the paper regarding GANs?\",\"answer\":\"To interpret and manipulate the latent space of GANs using a supervised method that maps latent variables to meaningful attributes.\"},{\"question\":\"How does the proposed approach incorporate human prior knowledge?\",\"answer\":\"By using labels from the input data to define a semantic label space and learning a linear mapping between latent space and these label variables.\"},{\"question\":\"Why is orthogonality regularization used in the method?\",\"answer\":\"To penalize similarity among mapping coefficients, alleviating unwanted effects caused by high correlations in the label space that would otherwise entangle attribute changes.\"}]","Interpreting the Latent Space of Generative Adversarial Networks using Supervised Learning | PDF",1787825929,15,{"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},"interpreting-the-latent-space-of-generative-adversarial-networks-using-supervised-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/interpreting-the-latent-space-of-generative-adversarial-networks-using-supervised-learning/150705/",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","2026-08-30","2026-08-27",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},"What is the main goal of the paper regarding GANs?","Question",{"text":76,"@type":77},"To interpret and manipulate the latent space of GANs using a supervised method that maps latent variables to meaningful attributes.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does the proposed approach incorporate human prior knowledge?",{"text":81,"@type":77},"By using labels from the input data to define a semantic label space and learning a linear mapping between latent space and these label variables.",{"name":83,"@type":74,"acceptedAnswer":84},"Why is orthogonality regularization used in the method?",{"text":85,"@type":77},"To penalize similarity among mapping coefficients, alleviating unwanted effects caused by high correlations in the label space that would otherwise entangle attribute changes.","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"]