BEGIN:VCALENDAR

VERSION:2.0

PRODID:-//wp-events-plugin.com//6.6.4.4//EN

TZID:Asia/Jerusalem

X-WR-TIMEZONE:Asia/Jerusalem
BEGIN:VEVENT

UID:0-801@aerospace.technion.ac.il

DTSTART;TZID=Asia/Jerusalem:20240404T110000

DTEND;TZID=Asia/Jerusalem:20240404T120000

DTSTAMP:20240325T074239Z

URL:https://aerospace.technion.ac.il/events/simplified-pomdp-algorithms-wi
 th-performance-guarantees/

SUMMARY:Simplified POMDP Algorithms with Performance Guarantees
DESCRIPTION:Lecturer:Moran Barenboim\n Faculty:TASP\n Institute:Technion 
 – Israel Institute of Technology\n Location:Zoom\n Zoom: https://technio
 n.zoom.us/j/91734030277?pwd=TUNEc3BkNzhLNXdxQVBoVGhOc0hwZz09\n Abstract: \
 n\n\n\nAutonomous agents operating in real-world scenarios frequently enco
 unter uncertainty and make decisions based on incomplete information. This
  challenge can be structured mathematically through the lens of partially 
 observable Markov decision processes (POMDPs). While POMDPs offer a robust
  framework for planning under uncertainty\, finding an optimal plan for a 
 POMDP can be computationally intensive and is feasible only for simpler ta
 sks. In response\, the last two decades have witnessed the rise of approxi
 mate algorithms\, like tree search and sample-based approaches\, as leadin
 g solutions for tackling more complex POMDP problems. Despite their effect
 iveness\, these algorithms typically offer only probabilistic guarantees o
 r\, in some cases\, no formal guarantees at all.\nIn our research\, we hav
 e focused on addressing these limitations by developing a range of simplif
 ied algorithms with formal\, deterministic guarantees. These simplified al
 gorithms operate on a selected subset of the state and observation spaces\
 , commonly considered in state-of-the-art algorithms\, while providing mat
 hematical guarantees and computational efficiencies compared to the non-si
 mplified algorithm. Initially\, we focused on a belief-dependent reward fr
 amework\, simplifying the reward calculation by narrowing down the observa
 tion space. Then\, we have applied a simplification to the state space in 
 the context of hybrid-belief and data-association aware POMDPs\, which oth
 erwise may grow exponentially. Ultimately\, we extended our approach to a 
 broad POMDP framework\, simplifying both state and observation spaces\, an
 d providing deterministic guarantees with respect to the optimal solution.
 \n\n\n\n\n Details: \n 
CATEGORIES:Seminars
LOCATION:Zoom

END:VEVENT

BEGIN:VTIMEZONE

TZID:Asia/Jerusalem

X-LIC-LOCATION:Asia/Jerusalem

BEGIN:DAYLIGHT

DTSTART:20240329T030000

TZOFFSETFROM:+0200

TZOFFSETTO:+0300

TZNAME:IDT

END:DAYLIGHT

END:VTIMEZONE
END:VCALENDAR