[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-514080-105":3,"detail-sidebar-cat-0-en-105":85,"doc-detail-514080-en":134},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":78,"head_meta":80,"extra_data":82,"updated_unix":84},105,"en","mixed-variable-bayesian-optimization-with-frequency-modulated-kernels","Mixed Variable Bayesian Optimization with Frequency Modulated Kernels","","Bayesian optimization benefits from Gaussian Process surrogate models, yet on mixed-variable spaces non-GP surrogates are often used due to limited kernel choices for complex cross-type dependencies. This paper introduces frequency modulated (FM) kernels that combine distances on continuous variables with a graph Fourier spectrum from discrete variables. Conditions are specified and proven to ensure positive definiteness and similarity-measure behavior. Experiments show improved sample efficiency over GP BO baselines and strong competitors, including Regularized Evolution and BOHB, even with fewer evaluations.",{"@graph":14,"@context":77},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/mixed-variable-bayesian-optimization-with-frequency-modulated-kernels/514080/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/mixed-variable-bayesian-optimization-with-frequency-modulated-kernels/514080.png","ImageObject",300,407,{"name":42,"@type":43},"Levi","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-15","2026-10-06",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",8,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69,73],{"name":60,"@type":61,"acceptedAnswer":62},"Why are non-Gaussian-process surrogates common in mixed variable Bayesian optimization?","Question",{"text":63,"@type":64},"Because on mixed-variable spaces it is difficult to define kernels that flexibly model dependencies across different types of variables.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How do frequency modulated (FM) kernels work in the proposed approach?",{"text":68,"@type":64},"They use distances on continuous variables to modulate the graph Fourier spectrum derived from discrete variables.",{"name":70,"@type":61,"acceptedAnswer":71},"What guarantees do the paper’s conditions provide for FM kernels?",{"text":72,"@type":64},"The paper specifies and proves conditions that make FM kernels positive definite and ensure they exhibit the similarity-measure behavior.",{"name":74,"@type":61,"acceptedAnswer":75},"What experimental results support the effectiveness of BO-FM?",{"text":76,"@type":64},"On synthetic and hyperparameter optimization tasks, BO-FM consistently outperforms competitors, and on joint optimization of neural architectures and SGD hyperparameters it outperforms Regularized Evolution and BOHB, even with fewer effective evaluations.","https://schema.org",{"og:url":32,"og:type":79,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":81,"canonical":32},"index,follow",{"doc_id":83,"site_id":7},514080,1791424201,{"code":4,"msg":86,"data":87},"success",[88,92,96,100,105,110,115,118,123,126,130],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":97,"show_sort_weight":98,"slug":99},"Exam",70,"exam",{"id":101,"doc_module":4,"doc_module_name":25,"category_name":102,"show_sort_weight":103,"slug":104},5,"Comic",60,"comic",{"id":106,"doc_module":4,"doc_module_name":25,"category_name":107,"show_sort_weight":108,"slug":109},6,"Technology",50,"technology",{"id":111,"doc_module":4,"doc_module_name":25,"category_name":112,"show_sort_weight":113,"slug":114},7,"Healthcare",40,"healthcare",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":116,"slug":117},30,"research-report",{"id":119,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":121,"slug":122},9,"Religion & Spirituality",20,"religion-spirituality",{"id":121,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":121,"slug":125},"World Cup","world-cup",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":127,"slug":129},10,"Lifestyle","lifestyle",{"id":131,"doc_module":4,"doc_module_name":25,"category_name":132,"show_sort_weight":101,"slug":133},19,"General","general",{"code":4,"msg":86,"data":135},{"doc_id":83,"user_id":136,"nickname":42,"user_avatar":137,"doc_module":4,"category_id":55,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":138,"file_id":139,"file_url":140,"file_type":141,"file_size":142,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":143,"language":144,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":145,"faqs":146,"seo_title":147,"seo_description":12,"update_tm":148,"read_time":149,"preview_page_count":33,"preview_urls":150},7971461740909,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","Mixed Variable Bayesian Optimization with Frequency Modulated Kernels  \nChangyong Oh 1 Efstratios Gavves 1 Max Welling 1,2  \n1Informatics Institute, University of Amsterdam, Amsterdam, The Netherlands  \n2 Qualcomm AI Research Netherlands, Amsterdam, The Netherlands  \nAbstract  \nThe sample efﬁciency of Bayesian optimization (BO) is often boosted by Gaussian Process (GP) surrogate models. However, on mixed variable spaces, surrogate models other than GPs are prevalent, mainly due to the lack of kernels which can model complex dependencies across different types of variables. In this paper, we propose the frequency modulated (FM) kernel ﬂexibly modeling dependencies among different types of variables, so that BO can enjoy the further improved sample efﬁciency. The FM kernel uses distances on continuous variables to modulate the graph Fourier spectrum derived from discrete variables. However, the frequency modulation does not always deﬁne a kernel with the similarity measure behavior which returns higher values for pairs of more similar points. Therefore, we specify and prove conditions for FM kernels to be positive deﬁnite and to exhibit the similarity measure behavior. In experiments, we demonstrate the improved sample efﬁciency of GP BO using FM kernels (BO-FM) .  \nOn synthetic problems and hyperparameter optimization problems, BO-FM outperforms competitors consistently. Also, the importance of the frequency modulation principle is empirically demonstrated on the same problems. On joint optimization of neural architectures and SGD hyperparameters, BO-FM outperforms competitors including Regularized evolution (RE) and BOHB. Remarkably, BO-FM performs better even than RE and BOHB using three times as many evaluations.  \n1 INTRODUCTION  \nBayesian optimization has found many applications ranging from daily routine level tasks of ﬁnding a tasty cookie  \nrecipe [Solnik et al., 2017] to sophisticated hyperparameter optimization tasks of machine learning algorithms (e.g. Alpha-Go [Chen et al., 2018]) . Much of this success is attributed to the ﬂexibility and the quality of uncertainty quantiﬁcation of Gaussian Process (GP)-based surrogate models [Snoek et al., 2012, Swersky et al., 2013, Oh et al., 2018] .  \nDespite the superiority of GP surrogate models, as compared to non-GP ones, their use on spaces with discrete structures (e.g., chemical spaces [Reymond and Awale, 2012], graphs and even mixtures of different types of spaces) is still application-speciﬁc [Kandasamy et al., 2018, Korovina et al., 2019] . The main reason is the difﬁculty ofdeﬁning kernels ﬂexible enough to model dependencies across different types of variables. On mixed variable spaces which consist of different types of variables including continuous, ordinal and nominal variables, current BO approaches resort to nonGP surrogate models, such as simple linear models or linear models with manually chosen basis functions [Daxberger et al., 2019] . However, such linear approaches are limited because they may lack the necessary model capacity.  \nThere is much progress on BO using GP surrogate models (GP BO) for continuous, as well as for discrete variables. However, for mixed variables it is not straightforward how to deﬁne kernels ,which can model dependencies across different types of variables. To bridge the gap, we propose frequency modulation which uses distances on continuous variables to modulate the frequencies of the graph spectrum [Ortega et al., 2018] where the graph represents the discrete part of the search space [Oh et al., 2019] .  \nA potential problem in the frequency modulation is that it does not always deﬁne a kernel with the similarity measure behavior [Vert et al., 2004] . That is, the frequency modulation does not necessarily deﬁne a kernel that returns higher values for pairs of more similar points. Formally, fora stationary kernel k(x; y) = s(x 􀀀 y), s should be decreasing [Remes et al., 2017] . In order to guarantee the similarity measure behavio","cbCaikSVXeAk79ud","https://ap.wps.com/l/cbCaikSVXeAk79ud","pdf",568850,11,"English","# Abstract\n# 1 Introduction\n# 2 Preliminaries","[{\"question\":\"Why are non-Gaussian-process surrogates common in mixed variable Bayesian optimization?\",\"answer\":\"Because on mixed-variable spaces it is difficult to define kernels that flexibly model dependencies across different types of variables.\"},{\"question\":\"How do frequency modulated (FM) kernels work in the proposed approach?\",\"answer\":\"They use distances on continuous variables to modulate the graph Fourier spectrum derived from discrete variables.\"},{\"question\":\"What guarantees do the paper’s conditions provide for FM kernels?\",\"answer\":\"The paper specifies and proves conditions that make FM kernels positive definite and ensure they exhibit the similarity-measure behavior.\"},{\"question\":\"What experimental results support the effectiveness of BO-FM?\",\"answer\":\"On synthetic and hyperparameter optimization tasks, BO-FM consistently outperforms competitors, and on joint optimization of neural architectures and SGD hyperparameters it outperforms Regularized Evolution and BOHB, even with fewer effective evaluations.\"}]","Mixed Variable Bayesian Optimization with Frequency Modulated Kernels | PDF",1791296022,28,[151,155,159,163],{"page_no":22,"url":152,"width":153,"height":154},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/87a77afbd95ff359f17d7fe15f2242be.webp",983,1307,{"page_no":26,"url":156,"width":157,"height":158},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/003e7fb7205ef00a9c63ff2170eace9c.webp",982,1306,{"page_no":30,"url":160,"width":161,"height":162},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/578cf9a2c03476a59273d2dd2e0fd3be.webp",981,1304,{"page_no":33,"url":164,"width":165,"height":166},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/486ccfed5e20f1eaf6e198f858f62576.webp",980,1303]