BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Europe/London BEGIN:DAYLIGHT DTSTART:20240331T020000 TZOFFSETFROM:+0000 TZOFFSETTO:+0100 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3 TZNAME:BST END:DAYLIGHT BEGIN:STANDARD DTSTART:20241027T010000 TZOFFSETFROM:+0100 TZOFFSETTO:+0000 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10 TZNAME:GMT END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20240517T112618Z UID:D6FDB0AB-667B-4566-ABEC-E67478FD4DC1 DTSTART;TZID=Europe/London:20240514T133000 DTEND;TZID=Europe/London:20240514T143000 DESCRIPTION:IEEE Aerospace and Electronic Systems Society - co-sponsored by AESS Boston (Chair Dr. Francesca Scire-Scappuzzo) and AESS London (Chair Dr. Julien Le Kernec)\n\nSeminar: Via formal verification\nAgenda: (Please note that since this is taking place in London\, for USA remote attendees the event will be a lunch time event 1:30PM - 2:30PM\n\n6:00-6:30 - In-Pe rson Networking\n6:30-7:30 - WebEx Presentation\n7:30-8:00 - In-Person Net working\n\nSpeaker: Prof. Alessio Lomuscio\, PhD\, Imperial College\, Lond on\, UK. Safe AI Lab. Royal Academy Engineering Chair in Emerging Technolo gies.\n\nA major challenge in deploying ML-based systems\, such as ML-base d computer vision\, is the inherent difficulty in ensuring their performan ce in the operational design domain. The standard approach consists in ext ensively testing models for inputs. However\, testing is inherently limite d in coverage\, and it is expensive in several domains. Novel verification methods provide guarantees that a neural model meets its specifications i n dense neighborhood of selected inputs. For example\, by using verificati on methods we can establish whether a model is robust with respect to infi nitely many lighting perturbations\, or particular noise patterns in the v icinity to an input. Verification methods can also be tailored to specific ations in the latent space and establish the robustness of models against semantic perturbations not definable in the input space (3D pose changes\, background changes\, etc). Additionally\, verification methods can be pai red with learning to obtain robust learning methods capable of generating models inherently more robust than those that may be derived with standard methods. In this presentation I will succinctly cover the key theoretical results leading to some of the present ML verification technology\, illus trate the resulting toolsets and capabilities\, and describe some of the u se cases developed with our colleagues at Boeing Research\, including cent erline distance estimation\, object detection\, and runway detection. I wi ll argue that verification and robust learning can be used to obtain model s that are inherently more robust and better understood than present learn ing and testing approaches\, thereby unlocking the deployment of applicati ons in the industry.\n\nSpeaker(s): Prof. Alessio Lomuscio\, PhD\, Imperia l College\, London\, UK. \, \n\nQinetiQ UK – 78\, St. James’ Street\, SW1A 1JB\, England\, United Kingdom\, Virtual: https://events.vtools.ieee. org/m/413287 LOCATION:QinetiQ UK – 78\, St. James’ Street\, SW1A 1JB\, England\, Uni ted Kingdom\, Virtual: https://events.vtools.ieee.org/m/413287 ORGANIZER:fscire@icloud.com SEQUENCE:36 SUMMARY:"Towards the assurance of AI-based systems" URL;VALUE=URI:https://events.vtools.ieee.org/m/413287 X-ALT-DESC:Description: <br /><p>IEEE Aerospace and Electronic Systems Soci ety - co-sponsored by AESS Boston (Chair Dr. Francesca Scire-Scappuzzo) an d AESS London (Chair Dr. Julien Le Kernec)</p>\n<p><strong>Seminar</strong >:&nbsp\; Via formal verification</p>\n<div><strong>Agenda: (Please note t hat since this is taking place in London\, for USA remote attendees the ev ent will be a lunch time event 1:30PM - 2:30PM</strong></div>\n<div>&nbsp\ ;</div>\n<div>\n<div>6:00-6:30 - In-Person Networking</div>\n<div>6:30-7:3 0 - WebEx Presentation</div>\n<div>7:30-8:00 - In-Person Networking</div>\ n<div>&nbsp\;</div>\n<p><strong>Speaker</strong>: Prof. Alessio Lomuscio\, PhD\, Imperial College\, London\, UK. Safe AI Lab. Royal Academy Engineer ing Chair in Emerging Technologies.</p>\n<p>A major challenge in deploying ML-based systems\, such as ML-based computer vision\, is the inherent dif ficulty in ensuring their performance in the operational design domain. Th e standard approach consists in extensively testing models for inputs. How ever\, testing is inherently limited in coverage\, and it is expensive in several domains. Novel verification methods provide guarantees that a neur al model meets its specifications in dense neighborhood of selected inputs . For example\, by using verification methods we can establish whether a m odel is robust with respect to infinitely many lighting perturbations\, or particular noise patterns in the vicinity to an input. Verification metho ds can also be tailored to specifications in the latent space and establis h the robustness of models against semantic perturbations not definable in the input space (3D pose changes\, background changes\, etc). Additionall y\, verification methods can be paired with learning to obtain robust lear ning methods capable of generating models inherently more robust than thos e that may be derived with standard methods. In this presentation I will s uccinctly cover the key theoretical results leading to some of the present ML verification technology\, illustrate the resulting toolsets and capabi lities\, and describe some of the use cases developed with our colleagues at Boeing Research\, including centerline distance estimation\, object det ection\, and runway detection. I will argue that verification and robust l earning can be used to obtain models that are inherently more robust and b etter understood than present learning and testing approaches\, thereby un locking the deployment of applications in the industry.</p>\n</div> END:VEVENT END:VCALENDAR