BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260525T225216EDT-1755eBHwx4@132.216.98.100
DTSTAMP:20260526T025216Z
DESCRIPTION:Abstract\n\nDecision-making is a fundamental problem in the mod
 ern world. Technology has developed to a level where automated decision-ma
 king is used even in safety-critical systems such as self-driving cars and
  industrial gas turbines. The design and operation of such systems often r
 equires decision-making without full-knowledge or information\, i.e.\, dec
 ision-making under uncertainty.\n\nUncertainty can manifest itself in many
  ways. Notable examples include non-reliable and/or delayed information. T
 his thesis investigates methods to advance and improve the reliability of 
 artificial intelligence led decision-making systems under these forms of u
 ncertainty. This thesis covers both sequential decision-making and predict
 ive systems. These are investigated on the backdrop of two industrial appl
 ication spaces: food retail and cyber-physical systems.\n\nFirst\, this th
 esis develops novel algorithms for multi-armed bandits and sequential deci
 sion-making. We present the first practical computation and indexing metho
 d for the optimal policy for Bernoulli multi-armed bandits\, which was pre
 viously considered intractable for several decades. Furthermore\, we provi
 de the optimal policy for Bernoulli bandits with delayed decision outcomes
 . We benchmark and gauge existing popular algorithms and showcase how perf
 ormance deteriorates significantly in the presence of delay. We then gener
 alize the concepts to non-Bernoulli distributions with delay.\n\nTo exploi
 t sequential decision-making in a practical application\, we build a simul
 ator to serve as a sandbox for experimentation to reduce food waste in foo
 d retail. We present a flexible framework capable of simulating a variety 
 of food retail entities and their interactions. Each entity is controllabl
 e by reinforcement learning agents. We demonstrate how combining simulatio
 n with reinforcement learning can effectively reduce food waste and increa
 se profits relative to a baseline.\n\nFinally\, the thesis investigates an
 d provides methodologies for building more robust predictive systems in th
 e presence of information uncertainty. Many industrial problems require de
 cision-making with limited information or potentially unreliable informati
 on. In collaboration with Siemens Energy as industrial partner\, we develo
 p machine learning predictors used for designing aeroderivative gas turbin
 es as complex cyber-physical systems. We also provide a methodology for de
 ploying such machine learning predictors to existing resource-constrained 
 control hardware.\n\nIn conclusion\, this thesis provides novel decision-m
 aking techniques for various forms of uncertainty by exploiting both theor
 etical and practical results across different application domains.\n
DTSTART:20240822T143000Z
DTEND:20240822T163000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Sebastian Pilarski – Artificial Intelligence Driven 
 Decision-Making Under Uncertainty
URL:/ece/channels/event/phd-defence-sebastian-pilarski
 -artificial-intelligence-driven-decision-making-under-uncertainty-358491
END:VEVENT
END:VCALENDAR