[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-128411-en":3,"doc-seo-128411-105":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},128411,8796095027276,"Valentina","https://avatar.qwps.com/avatar/d3BzX2FwX3Rlc3RfMjUxMTI2XzAxODA=",8,"Research & Report","Attributes Shape the Embedding Space of Face Recognition Models - ML Research Paper","Face Recognition (FR) advances rely on deep neural networks and margin-based triplet losses that map face images into high-dimensional embeddings. Training typically treats supervision as identity-only contrastive signals, yet a multiscale geometric structure emerges in the learned space. The work proposes a geometric description of how FR models depend on or remain invariant to interpretable facial and image attributes, and introduces a physics-inspired alignment metric for evaluation on controlled and augmented settings.","POLITECNICO DI TORINO Repository ISTITUZIONALE  \nAttributes Shape the Embedding Space of Face Recognition Models  \nOriginal  \nAttributes Shape the Embedding Space of Face Recognition Models / Vaccarino, Francesco; Nurisso, Marco; Leroy, Pierrick; Mastropietro, Antonio. -ELETTRONICO. -267:(2025), pp. 33960-33983. ( 42th International Conference on Machine Learning Vancouver, Canada 13-19 July 2025) .  \nAvailability:  \nThis version is available at: 11583/3006098 since: 2025-12-23T08:46:58Z  \nPublisher: PMLR  \nPublished DOI:  \nTerms of use:  \nThis article is made available under terms and conditions as specified in the corresponding bibliographic description in the repository  \nPublisher copyright  \n(Article begins on next page)  \n21 February 2026  \nAttributes Shape the Embedding Space of Face Recognition Models  \nPierrick Leroy* 1 Antonio Mastropietro* 2 Marco Nurisso* 1 3 Francesco Vaccarino 1  \nAbstract  \nFace Recognition (FR) tasks have made significant progress with the advent of Deep Neural Networks, particularly through margin-based triplet losses that embed facial images into highdimensional feature spaces. During training, these contrastive losses focus exclusively on identity information as labels. However, we observe amultiscale geometric structure emerging in the embedding space, influenced by interpretable facial (e.g., hair color) and image attributes (e.g., contrast) . We propose a geometric approach to describe the dependence or invariance of FR models to these attributes and introduce a physicsinspired alignment metric. We evaluate the proposed metric on controlled, simplified models and widely used FR models fine-tuned with synthetic data for targeted attribute augmentation. Our findings reveal that the models exhibit varying degrees of invariance across different attributes, providing insight into their strengths and weaknesses and enabling deeper interpretability. Code available here: [https://github.com/mantonios107/attrs](https://github.com/mantonios107/attrs)fr-embs.  \n1. Introduction  \nDeep Learning (DL) models have achieved state-of-the-art performance for Face Recognition (FR) tasks, due to increased availability of curated datasets (Deng et al., 2019b) and improved recognition architectures and losses (Denget al., 2022) . In the challenging open-set scenario, novel face identities can appear at testing time, hence the problem is framed as a metric learning task (Liu et al., 2017) . There are three main innovation to reach outstanding FR results in this scenario. The first consists of learning an embedding  \n*Equal contribution 1Department of Mathematical Sciences, Politecnico di Torino, Turin, Italy 2Department of Computer Science, University of Pisa, Pisa, Italy 3 CENTAI Institute, Turin, Italy. Correspondence to: Antonio Mastropietro \u003Canto[nio.mastropietro@di.unipi.it](nio.mastropietro@di.unipi.it) >.  \nProceedings of the 42 nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025 . Copyright 2025 by the author(s) .  \nspace representing images such that pictures of the same identity are closer in the embedding domain (Chopra et al., 2005) . Second, the usage of a triplet-loss, as in FaceNet (Schroff et al., 2015), where anchors and negative samples are compared to a target image to obtain the gradient driving the embedding map training (Sankaranarayanan et al., 2016) . The third consists in equipping the embedding space of an angular dissimilarity, so that it can be assimilated to a hypersphere, e.g., AM-Softmax (Wang et al., 2018a), SphereFace (Liu et al., 2017), CosFace (Wang et al., 2018b), ArcFace (Deng et al., 2019a), AdaFace (Kim et al., 2022) .  \nHowever, our understanding and interpretation of the structure of the embedding space learned by an FR model remain significantly underdeveloped. Unlike interpretable attributes, deep face representations exist in a highdimensional space, making it difficult to discern the specific properties they encode (O’Toole et al., 2018)","cbCaie5tkhcZLN1w","https://ap.wps.com/l/cbCaie5tkhcZLN1w","pdf",4001383,1,25,"English","en",105,"# Abstract\n# Introduction\n## Metric learning and embedding design\n## Angular dissimilarity and hyperspherical embeddings\n## Limits of interpretability in embedding spaces\n## Prior work on analyzing and recovering attributes","[{\"question\":\"How do the proposed ideas relate to triplet-loss training in face recognition?\",\"answer\":\"The approach starts from the observation that margin-based triplet losses embed faces into a feature space while supervision focuses on identity. It then studies additional geometric structure beyond identity-only training signals.\"},{\"question\":\"What does the paper say about the embedding space’s structure with respect to facial attributes?\",\"answer\":\"It reports that a multiscale geometric structure appears in the embedding space and is influenced by interpretable facial and image attributes, such as hair color and contrast.\"},{\"question\":\"How is attribute dependence or invariance evaluated?\",\"answer\":\"The paper introduces a physics-inspired alignment metric and evaluates it on controlled simplified models and fine-tuned FR models using synthetic data for attribute augmentation.\"}]","Attributes Shape the Embedding Space of Face Recognition Models - ML Research Paper | PDF",1785947364,63,{"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},"attributes-shape-the-embedding-space-of-face-recognition-models-ml-research-paper","",{"@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/attributes-shape-the-embedding-space-of-face-recognition-models-ml-research-paper/128411/",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","2026-08-23","2026-08-05",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},"How do the proposed ideas relate to triplet-loss training in face recognition?","Question",{"text":76,"@type":77},"The approach starts from the observation that margin-based triplet losses embed faces into a feature space while supervision focuses on identity. It then studies additional geometric structure beyond identity-only training signals.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What does the paper say about the embedding space’s structure with respect to facial attributes?",{"text":81,"@type":77},"It reports that a multiscale geometric structure appears in the embedding space and is influenced by interpretable facial and image attributes, such as hair color and contrast.",{"name":83,"@type":74,"acceptedAnswer":84},"How is attribute dependence or invariance evaluated?",{"text":85,"@type":77},"The paper introduces a physics-inspired alignment metric and evaluates it on controlled simplified models and fine-tuned FR models using synthetic data for attribute augmentation.","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},[94,98,102,106,111,116,121,124,129,132,136],{"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":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":46,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":133,"doc_module":4,"doc_module_name":46,"category_name":134,"show_sort_weight":133,"slug":135},10,"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":46,"category_name":138,"show_sort_weight":107,"slug":139},19,"General","general"]