BEGIN:VCALENDAR
VERSION:2.0
X-WR-CALNAME;VALUE=TEXT:Cautious Bandits
PRODID:-//Harvard events data//EN
BEGIN:VEVENT
UID:event_1618795_0
SUMMARY:Cautious Bandits
DESCRIPTION:<p>	<strong>Presenter</strong>:  Andy Haupt<br><strong>Topic</strong>:  Cautious Bandits</p><p>	<!--break--></p><h2>	<strong>Fall 2021</strong></h2><p>	The EconCS Group holds an Economics and Computer Science research seminar each semester.</p><p>	Fall 2021 meetings are held at 1 - 2:30 PM on Fridays. Seminar Coordinators for Fall '21 are <a href="mailto:saisr@g.harvard.edu">Srivatsa R Sai</a> and <a href="mailto:dhalpern@g.harvard.edu">Daniel Halpern</a>.</p><p>	<strong>Abstract: </strong>We introduce and characterize revealed risk preferences of bandit algorithms. An algorithm for the stochastic bandit problem is risk averse if for any fixed noise levels and time, there is a reward difference such that the algorithm chooses a less risky arm over a higher expected reward risky arm, with high probability in time. We experimentally find that several classical adversarial and stochastic bandit algorithms (eps-Greedy, UCB, EXP3) and prove that eps-Greedy is risk-averse. We discuss implications for the separation of learning and deployment of reinforcement learning algorithms and discuss extensions of our statement to mean-based bandit algorithms (Braverman et al. 2018) and to multi-agent environments.</p><p>	 </p>
LOCATION:
STATUS:CONFIRMED
DTSTART:20211015T170000Z
DTEND:20211015T183000Z
END:VEVENT
END:VCALENDAR