{"id":10456,"date":"2026-02-13T07:02:33","date_gmt":"2026-02-13T07:02:33","guid":{"rendered":"https:\/\/mailitics.com\/index.php\/2026\/02\/13\/2602-11406\/"},"modified":"2026-02-13T07:02:33","modified_gmt":"2026-02-13T07:02:33","slug":"2602-11406","status":"publish","type":"post","link":"https:\/\/mailitics.com\/index.php\/2026\/02\/13\/2602-11406\/","title":{"rendered":"The Cost of Learning under Multiple Change Points"},"content":{"rendered":"<p>    The Cost of Learning under Multiple Change Points<br \/>\n \t<BR><br \/>\n<BR><\/BR><br \/>\n    <!-- no image --><br \/>\n \t<BR><br \/>\n<BR><\/BR><\/p>\n<div>arXiv:2602.11406v1 Announce Type: new<br \/>\nAbstract: We consider an online learning problem in environments with multiple change points. In contrast to the single change point problem that is widely studied using classical &#8220;high confidence&#8221; detection schemes, the multiple change point environment presents new learning-theoretic and algorithmic challenges. Specifically, we show that classical methods may exhibit catastrophic failure (high regret) due to a phenomenon we refer to as endogenous confounding. To overcome this, we propose a new class of learning algorithms dubbed Anytime Tracking CUSUM (ATC). These are horizon-free online algorithms that implement a selective detection principle, balancing the need to ignore &#8220;small&#8221; (hard-to-detect) shifts, while reacting &#8220;quickly&#8221; to significant ones. We prove that the performance of a properly tuned ATC algorithm is nearly minimax-optimal; its regret is guaranteed to closely match a novel information-theoretic lower bound on the achievable performance of any learning algorithm in the multiple change point problem. Experiments on synthetic as well as real-world data validate the aforementioned theoretical findings.<\/div>\n<p> \t<BR><br \/>\n <BR><\/BR><br \/>\n    Tomer Gafni, Garud Iyengar, Assaf Zeevi<br \/>\n \t<BR><br \/>\n<BR><\/BR><br \/>\n<a href=\"https:\/\/arxiv.org\/abs\/2602.11406\">Go to original source<\/a><br \/>\n \t<BR><br \/>\n <BR><\/BR><\/p>\n","protected":false},"excerpt":{"rendered":"<p>The Cost of Learning under Multiple Change Points arXiv:2602.11406v1 Announce Type: new Abstract: We consider an online learning problem in environments with multiple change points. In contrast to the single change point problem that is widely studied using classical &#8220;high confidence&#8221; detection schemes, the multiple change point environment presents new learning-theoretic and algorithmic challenges. Specifically, [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[62,113,112],"tags":[1019,199,1000],"class_list":["post-10456","post","type-post","status-publish","format-standard","hentry","category-aimldsaimlds","category-cs-lg","category-stat-ml","tag-change","tag-learning","tag-multiple"],"_links":{"self":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts\/10456"}],"collection":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/comments?post=10456"}],"version-history":[{"count":0,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts\/10456\/revisions"}],"wp:attachment":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/media?parent=10456"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/categories?post=10456"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/tags?post=10456"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}