{"id":89,"date":"2005-06-18T22:40:45","date_gmt":"2005-06-19T04:40:45","guid":{"rendered":"\/?p=89"},"modified":"2005-06-18T22:41:32","modified_gmt":"2005-06-19T04:41:32","slug":"lower-bounds-for-learning-reductions","status":"publish","type":"post","link":"https:\/\/hunch.net\/?p=89","title":{"rendered":"Lower Bounds for Learning Reductions"},"content":{"rendered":"

Learning reductions<\/a> transform a solver of one type of learning problem into a solver of another type of learning problem. When we analyze these for robustness we can make statement of the form “Reduction R<\/em> has the property that regret r<\/em> (or loss) on subproblems of type A<\/em> implies regret at most f ( r )<\/em> on the original problem of type B<\/em>“.<\/p>\n

A lower bound for a learning reduction would have the form “for all reductions R<\/em>, there exists a learning problem of type B<\/em> and learning algorithm for problems of type A<\/em> where regret r<\/em> on induced problems implies at least<\/em> regret f ( r )<\/em> for B<\/em>“.<\/p>\n

The pursuit of lower bounds is often questionable because, unlike upper bounds, they do not yield practical algorithms. Nevertheless, they may be helpful as a tool for thinking about what is learnable and how learnable it is. This has already come up here<\/a> and here<\/a>.<\/p>\n

At the moment, there is no coherent theory of lower bounds for learning reductions, and we have little understanding of how feasible they are or which techniques may be useful in proving them. Here is a rough summary of what I know:<\/p>\n

    \n
  1. For structured prediction<\/a>, we have a partially worked out lower bound for all reductions using the structure to only carry single bits<\/em>. A proof for reductions using the structure in others ways seems tricky at the moment.<\/li>\n
  2. For Reinforcement learning<\/a> it may (this is unclear) be possible to prove a lower bound showing that prediction ability alone can not solve RL well.<\/li>\n
  3. There are various results which can be thought of as lower bounds for more limited families of reductions. One example is analyzing exactly how badly margin optimization can underperform for 0-1 loss when there is noise.<\/a><\/li>\n<\/ol>\n

    Overall, this is a moderately interesting direction of research which has not been much investigated. <\/p>\n","protected":false},"excerpt":{"rendered":"

    Learning reductions transform a solver of one type of learning problem into a solver of another type of learning problem. When we analyze these for robustness we can make statement of the form “Reduction R has the property that regret r (or loss) on subproblems of type A implies regret at most f ( r … <\/p>\n

    Continue reading “Lower Bounds for Learning Reductions”<\/span><\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[16,12],"tags":[],"class_list":["post-89","post","type-post","status-publish","format-standard","hentry","category-problems","category-reductions"],"_links":{"self":[{"href":"https:\/\/hunch.net\/index.php?rest_route=\/wp\/v2\/posts\/89","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hunch.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hunch.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hunch.net\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/hunch.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=89"}],"version-history":[{"count":0,"href":"https:\/\/hunch.net\/index.php?rest_route=\/wp\/v2\/posts\/89\/revisions"}],"wp:attachment":[{"href":"https:\/\/hunch.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=89"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hunch.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=89"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hunch.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=89"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}