BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//chikkutakku.com//RDFCal 1.0//EN
X-WR-CALDESC:GoogleカレンダーやiCalendar形式情報を共有シェ
 アしましょう。近所のイベントから全国のイベントま
 で今日のイベント検索やスケジュールを決めるならち
 っくたっく
X-WR-CALNAME:ちっくたっく
X-WR-TIMEZONE:UTC
BEGIN:VEVENT
SUMMARY:School of IT seminar slot (compulsory if scheduled)
DTSTART;VALUE=DATE-TIME:20250821T110000Z
DTEND;VALUE=DATE-TIME:20250821T115000Z
UID:196155585730
DESCRIPTION:Title: Meta-Learning the Intrinsic Reward Weighting in Curiosi
 ty-Driven RL\n\nAbstract: Reinforcement learning agents must find a balanc
 e between\nexploitation and exploration to maximise the cumulative sum of\
 nextrinsic rewards. However\, it is particularly challenging for agents\nt
 o explore effectively in sparse reward environments where feedback is\nrar
 e. Curiosity-driven algorithms can be used to encourage effective\nexplora
 tion. These algorithms generate an additional reward called the\nintrinsic
  reward that encourages agents to seek novel situations. The\nintrinsic re
 wards are combined with extrinsic rewards using a weighted\nsum\, where λ
  is the weighting for the intrinsic reward. However\, λ is\noften fine-tu
 ned for each new environment\, even when environments are\nsimilar. We pro
 pose a meta-learning approach that replaces the fixed λ\nparameter with a
  neural network that outputs λ values at each time\nstep. This network is
  trained using evolutionary strategies and can\ngeneralise across similar 
 environments without retraining. Our\napproach highlights the potential fo
 r reducing the need for\nfine-tuning λ across similar tasks.\n\nBiography
 : Batsi is a Master's student at the University of Cape Town\nwith interes
 ts in curiosity-driven reinforcement learning and\nmeta-reinforcement lear
 ning. His research focuses on improving the\nsample efficiency of reinforc
 ement learning algorithms.
LOCATION:CS2A
END:VEVENT
END:VCALENDAR