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:20241027T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RDATE:20251026T030000
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20250330T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.29366.field_data.0@www.open.diag.uniroma1.it
DTSTAMP:20260403T173641Z
CREATED:20250522T150219Z
DESCRIPTION:Abstract -- The rapid advancement of multimodal foundation mode
 ls in language and vision has opened new possibilities for autonomous syst
 ems. These models offer broad perceptual and reasoning capabilities\, but 
 integrating them into the design and deployment of autonomous agents remai
 ns a challenge. Their outputs often lack guarantees\, may misalign with do
 main-specific tasks\, and introduce uncertainty that can compromise safety
  and reliability. In this talk\, I explore how foundation models can be ad
 apted and refined to meet the stringent requirements of autonomy by combin
 ing them with tools from formal methods and uncertainty quantification. I 
 will discuss a method for automatically fine-tuning pre-trained language m
 odels for control tasks using formal specification-guided synthesis of ver
 ifiable automaton-based controllers\, without relying on human feedback. I
  will also present a framework for disentangling and mitigating uncertaint
 y in multimodal planning pipelines by separating perceptual and decision u
 ncertainty\, applying conformal prediction\, and introducing a formal-meth
 ods-driven quantification technique. These approaches\, validated in simul
 ated and real-world robotic scenarios\, show that principled adaptation of
  foundation models—modulo formal verification and calibrated uncertainty—c
 an significantly improve both task performance and safety guarantees.Bio -
 - Ufuk Topcu is a Professor in the Department of Aerospace Engineering and
  Engineering Mechanics at The University of Texas at Austin. He is a core 
 faculty member of the Oden Institute for Computational Engineering and Sci
 ences\, Texas Robotics\, and Machine Learning Laboratory\, and he directs 
 the Autonomous Systems Group. His research lies at the intersection of for
 mal methods\, reinforcement learning\, and control theory\, with a focus o
 n the theoretical andalgorithmic foundations for the design and verificati
 on of autonomous systems. He is a recipient of the IEEE Control System Soc
 iety Antonio Ruberti Young Researcher Prize.
DTSTART;TZID=Europe/Paris:20250527T103000
DTEND;TZID=Europe/Paris:20250527T103000
LAST-MODIFIED:20250522T165304Z
LOCATION:DIAG - Aula Magna
SUMMARY:Foundation Models Modulo Formal Methods and Uncertainty Quantificat
 ion - Ufuk Topcu
URL;TYPE=URI:http://www.open.diag.uniroma1.it/node/29366
END:VEVENT
END:VCALENDAR