{"id":1595,"date":"2026-11-11T13:01:04","date_gmt":"2026-11-11T13:01:04","guid":{"rendered":"https:\/\/controlsociety.org\/icc\/?page_id=1595"},"modified":"2026-11-11T13:01:04","modified_gmt":"2026-11-11T13:01:04","slug":"prof-shalabh-bhatnagar-indian-institute-of-science","status":"publish","type":"page","link":"https:\/\/controlsociety.org\/icc\/prof-shalabh-bhatnagar-indian-institute-of-science\/","title":{"rendered":"Prof. Shalabh Bhatnagar (Indian Institute of Science)"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong><span style=\"text-decoration: underline;\">Title:<\/span><\/strong> Zeroth-Order Optimization Under Noise: Taming Bias and Non-smoothness<\/p>\n\n\n\n<p class=\"has-text-align-left wp-block-paragraph\"><strong><span style=\"text-decoration: underline;\">Abstract:<\/span><\/strong> Many engineering and learning problems have objective functions that must be optimized without access to the gradient &#8212; we can only evaluate the objective, often with noise. Gradient-free methods estimate the slope from such evaluations, but every estimate is biased, and many objectives, such as absolute-value losses or ReLU networks, have kinks where the gradient is not defined.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This talk presents two ways of handling these difficulties, linked by one viewpoint: the algorithm is a stochastic approximation whose long-run behaviour is read off from a deterministic limiting dynamics. In the first part, for smooth objectives, bias is a design choice. A few more function measurements per step, combined through a shift-operator expansion, reduce the bias to arbitrarily high order in the perturbation size. Since the bias is not zero, the limiting dynamics is a differential inclusion, a set-valued perturbation of gradient descent, and the iterates converge to a neighbourhood of the minimizers that shrinks with the bias.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In the second part, we drop smoothness. The gradient is replaced by the set-valued Clarke generalized gradient. Two noisy evaluations along a random Gaussian direction estimate the gradient of a smoothed objective, which approaches the generalized gradient as the smoothing is reduced. A two-timescale projected scheme averages this estimate on the fast scale and descends along it on the slow scale. In outline, the iterates track a projected differential inclusion, compared with the smoothed dynamics by a Gr\u00f6nwall argument, and converge almost surely to a neighbourhood of the Clarke stationary points, the candidates for a minimum.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n<!--themify_builder_content-->\n<div id=\"themify_builder_content-1595\" data-postid=\"1595\" class=\"themify_builder_content themify_builder_content-1595 themify_builder tf_clear\">\n    <\/div>\n<!--\/themify_builder_content-->\n<!-- AddThis Advanced Settings generic via filter on the_content --><!-- AddThis Share Buttons generic via filter on the_content -->","protected":false},"excerpt":{"rendered":"<p>Title: Zeroth-Order Optimization Under Noise: Taming Bias and Non-smoothness Abstract: Many engineering and learning problems have objective functions that must be optimized without access to the gradient &#8212; we can only evaluate the objective, often with noise. Gradient-free methods estimate the slope from such evaluations, but every estimate is biased, and many objectives, such as [&hellip;]<!-- AddThis Advanced Settings generic via filter on get_the_excerpt --><!-- AddThis Share Buttons generic via filter on get_the_excerpt --><\/p>\n","protected":false},"author":4,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-1595","page","type-page","status-publish","hentry","has-post-title","has-post-date","has-post-category","has-post-tag","has-post-comment","has-post-author",""],"builder_content":"","_links":{"self":[{"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/pages\/1595","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/comments?post=1595"}],"version-history":[{"count":1,"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/pages\/1595\/revisions"}],"predecessor-version":[{"id":1596,"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/pages\/1595\/revisions\/1596"}],"wp:attachment":[{"href":"https:\/\/controlsociety.org\/icc\/wp-json\/wp\/v2\/media?parent=1595"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}