{"id":2688,"date":"2020-10-23T11:08:34","date_gmt":"2020-10-23T16:08:34","guid":{"rendered":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/?post_type=tribe_events&p=2688"},"modified":"2020-12-08T14:29:37","modified_gmt":"2020-12-08T20:29:37","slug":"virtual-lans-seminar-19","status":"publish","type":"tribe_events","link":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/event\/virtual-lans-seminar-19\/","title":{"rendered":"Virtual LANS Seminar"},"content":{"rendered":"

Seminar Title<\/strong>: Should We Be Perturbed About Deep Learning?
\nSpeaker<\/strong>: Desmond Higham, Professor of Numerical Analysis, The University of Edinburgh<\/p>\n

Date\/Time<\/strong>: December 9, 2020 \/ 11:00 AM – 12:00 PM
\nLocation<\/strong>: See meeting URL on the cels-seminars website<\/a> (requires Argonne login)<\/p>\n

\n

Description<\/strong>: Many commentators are asking whether current AI solutions are sufficiently robust, resilient, and trustworthy; and how such issues should be quantified and addressed. I believe that numerical analysts can contribute to the debate.\u00a0In part 1 of this talk I will look at the common practice of using low precision floating point formats to speed up computation time. I will focus on evaluating the softmax<\/u>\u00a0and log-sum-exp functions, which play an important role in many classification tools. Here, across widely used packages we see mathematically equivalent but computationally different formulations; these variations have been designed in an effort to avoid overflow and underflow. I will show that classical rounding error analysis gives insights into their floating-point accuracy, and suggests a method of choice. In part 2 I will look at a bigger picture question concerning sensitivity to adversarial attacks in deep learning. Adversarial attacks are deliberate, targeted perturbations to input data that have a dramatic effect on the output; for example, a traffic “Stop” sign on the roadside can be misinterpreted as a speed limit sign when minimal graffiti is added. The vulnerability of systems to such interventions raises questions around security, privacy and ethics, and there has been a rapid escalation of attack and defense strategies. I will consider a related higher-level question: under realistic assumptions, do adversarial examples always exist with high probability? I will also introduce and discuss the idea of a stealth attack: an undetectable, targeted perturbation to the trained network itself.<\/p>\n

Part 1 is joint work with Pierre Blanchard (ARM) and Nick Higham<\/u> (Manchester).<\/p>\n

Part 2 is joint work with Alexander Gorban<\/u> and Ivan Tyukin<\/u> (Leicester).<\/p>\n

Please note that the meeting URL for this event can be seen on the <\/em>cels-seminars website<\/a>, which requires an Argonne login.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

Seminar Title: Should We Be Perturbed About Deep Learning? Speaker: Desmond Higham, Professor of Numerical Analysis, The University of Edinburgh Date\/Time: December 9, 2020 \/ 11:00 AM – 12:00 PM Location: See meeting URL on the cels-seminars website (requires Argonne … Continue reading →<\/span><\/a><\/p>\n","protected":false},"author":43,"featured_media":0,"template":"","meta":{"_acf_changed":false,"_monsterinsights_skip_tracking":false,"_tribe_events_status":"","_tribe_events_status_reason":"","footnotes":"","_members_access_role":[],"_members_access_error":""},"tags":[],"tribe_events_cat":[2],"class_list":["post-2688","tribe_events","type-tribe_events","status-publish","hentry","tribe_events_cat-seminar","cat_seminar"],"acf":[],"_links":{"self":[{"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tribe_events\/2688","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tribe_events"}],"about":[{"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/types\/tribe_events"}],"author":[{"embeddable":true,"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/users\/43"}],"version-history":[{"count":2,"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tribe_events\/2688\/revisions"}],"predecessor-version":[{"id":2708,"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tribe_events\/2688\/revisions\/2708"}],"wp:attachment":[{"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/media?parent=2688"}],"wp:term":[{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tags?post=2688"},{"taxonomy":"tribe_events_cat","embeddable":true,"href":"https:\/\/wordpress.cels.anl.gov\/lans-seminars\/wp-json\/wp\/v2\/tribe_events_cat?post=2688"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}