[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82548-en":3,"doc-seo-82548-105":29,"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":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":13,"seo_description":14,"update_tm":27,"read_time":28},82548,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Generative Refinement for Low-Budget Black-Box Optimization","Black-box optimization enables optimizing objectives without gradient information, but becomes difficult when evaluations are expensive, noisy, failure-prone, or when high-performing solutions occupy thin, curved, or disconnected regions. Many generative-model methods sample from reward-aligned distributions, requiring many evaluations to align the sampler and making them impractical under tight budgets. SPARROW decouples the generative prior from reward by using a fixed structured proposal from an unconditional sampler plus rank-based guidance over an evaluated archive, offering robustness, convergence guarantees, and strong empirical results.","Generative Refinement for Low-Budget Black-Box  \nOptimization  \narXiv :2607 .0069 1v 1 [ cs .LG] 1 Jul 2026  \nEdouard R. Dufour  \nCVLab EPFL Lausanne, CH  \n[edouard.dufour@epfl.ch](edouard.dufour@epfl.ch)  \nPascal Fua  \nCVLab EPFL Lausanne, CH [pascal.fua@epfl.ch](pascal.fua@epfl.ch)  \nAbstract  \nBlack-box optimization is a fundamental science and engineering tool that makes it possible to optimize objectives without gradient information. Unfortunately, as it often requires many function evaluations, it can be challenging when each one is costly. This is especially true when the evaluation function is noisy or failure-prone, and when high-performing solutions are confined to thin, curved, or disconnected regions of the search space. Existing methods leveraging generative models to navigate these subspaces are built to sample from reward-aligned distributions. Asa result, they require a large number of evaluations to align their sampler effectively, making them impractical in low-budget settings. We propose SPARROW, an algorithm that completely decouples the generative prior from the reward signal.  \nSPARROW can use any sampler with a known corruption process and trained on unevaluated data, as a fixed, structured proposal operator. Optimization proceeds by rank-based guidance over an archive of evaluated candidates. SPARROW can navigate complex geometries, handle unreliable reward signals, and perform effective optimization under very low evaluation budgets. We provide asymptotic convergence guarantees over the sampler support and demonstrate strong empirical performance on problems with unreliable rewards and geometrically complex landscapes.  \n1 Introduction  \nMany problems in science and engineering require optimizing without access to gradient information, as in materials design [9], drug discovery [10, 11], and engineering simulation [5, 39] . This is referred to as Black-Box Optimization (BBO) . It is particularly challenging when evaluations are costly, limiting the evaluation budget to tens or hundreds of candidates rather than tens of thousands. The difficulty is magnified when the evaluation feedback is noisy or failure-prone and when good solutions lie on thin, curved or disconnected regions of the search space.  \nIn these settings, classical approaches struggle. Bayesian optimization (BO) methods such as Gaussian processes (GP) [27] are efficient but suffer when the feedback is unreliable. Evolutionary strategies (ES) such as CMA-ES [13] are more robust but typically need more feedback to get going. All make distributional assumptions that degrade with dimensionality and complexity of the search space, leading them to waste evaluations on infeasible candidates and failing to account for the geometry of the problem.  \nDiffusion [15, 30, 29] and flow matching [21] models have recently emerged as powerful tools for representing complex, high-dimensional distributions. This has motivated their use in BBO for settings where domain data is available. Such data is however often unlabeled with respect to the chosen objective, providing structural information but no objective signal. Existing methods target  \nPreprint.  \na reward-weighted distribution [18, 38, 34], which requires either a labeled training dataset or a large number of evaluations during sampling to accurately shape the sampler. Thus, the distributionlearning paradigm is at odds with the practical objective of identifying the best single solution under a limited budget.  \nWe propose SPARROW (Sequential Proposal via Archival Rank-based Refinement for Optimization under Weak feedback), a new BBO algorithm that decouples generative modeling from optimization. SPARROW uses a fixed, unconditional sampler as a proposal operator and requires only access to its corruption and sampling processes, regardless of internal structure. Optimization is driven by rankbased guidance over an archive of evaluated candidates, giving robustness to unreliable feedback. We provide a","cbCaiqwQ4SP1r0F0","https://ap.wps.com/l/cbCaiqwQ4SP1r0F0","pdf",736900,1,20,"English","en",105,"# Abstract\n# Introduction\n# Related Work\n## Classical Black-Box Optimization\n## Generative Models for Black-Box Optimization","[{\"question\":\"What challenge does low-budget black-box optimization address?\",\"answer\":\"It focuses on optimizing objectives without gradients when only tens or hundreds of costly evaluations are allowed, especially under noisy, failure-prone feedback and when good solutions lie in thin, curved, or disconnected search regions.\"},{\"question\":\"Why are existing generative-model approaches often impractical in low-budget settings?\",\"answer\":\"They typically learn or sample from reward-aligned distributions, which requires a large number of evaluations to accurately shape the sampler, creating high evaluation cost.\"},{\"question\":\"How does SPARROW decouple generative modeling from the reward signal?\",\"answer\":\"SPARROW uses an unconditional generative prior as a fixed structured proposal operator (based on a known corruption process), and performs optimization using rank-based guidance over an archive of evaluated candidates rather than retraining the proposal to the reward.\"}]",1784181464,50,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"generative-refinement-for-low-budget-black-box-optimization","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/generative-refinement-for-low-budget-black-box-optimization/82548/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What challenge does low-budget black-box optimization address?","Question",{"text":75,"@type":76},"It focuses on optimizing objectives without gradients when only tens or hundreds of costly evaluations are allowed, especially under noisy, failure-prone feedback and when good solutions lie in thin, curved, or disconnected search regions.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why are existing generative-model approaches often impractical in low-budget settings?",{"text":80,"@type":76},"They typically learn or sample from reward-aligned distributions, which requires a large number of evaluations to accurately shape the sampler, creating high evaluation cost.",{"name":82,"@type":73,"acceptedAnswer":83},"How does SPARROW decouple generative modeling from the reward signal?",{"text":84,"@type":76},"SPARROW uses an unconditional generative prior as a fixed structured proposal operator (based on a known corruption process), and performs optimization using rank-based guidance over an archive of evaluated candidates rather than retraining the proposal to the reward.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,126,129,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":28,"slug":113},6,"Technology","technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":21,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":21,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":106,"slug":136},19,"General","general"]