BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Date iCal//NONSGML kigkonsult.se iCalcreator 2.20.2//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:Eventi DIAG
BEGIN:VTIMEZONE
TZID:Europe/Paris
BEGIN:STANDARD
DTSTART:20211031T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20220327T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.23890.field_data.0@www.ugov-ricerca.uniroma1.it
DTSTAMP:20260406T010728Z
CREATED:20211103T065510Z
DESCRIPTION:A key difficulty in the deployment of AI solutions\, including 
 machine learning\, remains their inherent fragility and difficulty of cert
 ification. Formal verification has long been employed in the analysis and 
 debugging of traditional computer systems\, including hardware\, but its d
 eployment in the context of safety-critical AI-systems remains largely une
 xplored. In this talk I will summarise some of the contributions on verifi
 cation of neural systems from the Verification of Autonomous Systems Lab a
 t Imperial College London. I will focus on the issue of specifications and
  verifications for deep neural classifiers. After a discussion on specific
 ations\, I will introduce recent exact and approximate methods\, including
  MILP-based approaches\, linear relaxations\, and symbolic interval propag
 ation. I will showcase the resulting toolkits\, including Venus and Verine
 t\, and exemplify their use on scenarios developed within the DARPA projec
 t on Assured Autonomy. This will enable us to observe that the verificatio
 n of neural models with hundreds of thousands of nodes (corresponding to m
 illions of tuneable paramaters) is now feasible with further advances like
 ly to be achieved in the near future. Time permitting\, I will briefly dis
 cuss closely related ongoing work\, including verification of neural-symbo
 lic multi-agent systems (closed-loop AI systems combining neural and symbo
 lic components).
DTSTART;TZID=Europe/Paris:20211105T160000
DTEND;TZID=Europe/Paris:20211105T160000
LAST-MODIFIED:20211103T083955Z
LOCATION:DIAG\, Aula Magna
SUMMARY:Towards verifying AI systems based on deep Neural networks - Prof. 
 Alessio Lomuscio (Imperial College London\, UK) - Prof. Alessio Lomuscio (
 Imperial College London\, UK)
URL;TYPE=URI:http://www.ugov-ricerca.uniroma1.it/node/23890
END:VEVENT
END:VCALENDAR