DASH
Repository logo
Communities & Collections
Browse DASH
DASH Stories
Statistics
  1. Home
  2. Faculty of Arts and Sciences
  3. FAS Theses and Dissertations
  4. Approaches to Policy Advice From Multiple Teachers in Reinforcement Learning

Approaches to Policy Advice From Multiple Teachers in Reinforcement Learning

Open/View files
LI-SENIORTHESIS-2017.pdf (14.95 MB)
Date
July 14, 2017
Author(s)
The Harvard community has made this article openly available. Please share how this access benefits you.
Abstract
Reinforcement learning can be an arduous and expensive process, but agents can learn a task faster and better with the help of one or more teacher agents. In the advice model framework, a teacher instructs by giving advice to the student during its learning. In budgeted learning, the teacher can only dispense advice a limited number of times, which is more realistic in many domains and doesn't require the continual attention of the teacher to the student. This thesis explores how to extend the budgeted student-teacher framework to multiple teachers on budgets, framing for student agents as well as keeping the possibility of extending to human students, which would open up new possibilities in a plethora of domains. I evaluate the performance of multiple teachers in the Pac-Man domain. I show that it is possible for multiple teacher agents to outperform the single best teacher, that adding ``bad" teachers can have a negative effect on the student's learning, and that certain combinations of teaching heuristics with the number of teachers can be more beneficial to learning than others. I also present a simple model with promising learning results which leverages multiple individual teachers rather than aggregating all of them into a single entity.
Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material (LAA), as set forth at Terms of Service
URI
http://nrs.harvard.edu/urn-3:HUL.InstRepos:38811482
Subjects

Computer Science

Collection
FAS Theses and Dissertations
Metrics
Repository logo
  • CONTACT US
  • OSRDS
  • ABOUT DASH
  • DEPOSITING TO DASH
  • TERMS OF USE
  • REQUEST A WAIVER
  • TAKE DOWN POLICY
  • NAME CHANGE POLICY
  • HARVARD OA POLICIES
  • DASH SUBMITTER AGREEMENT
  • Accessibility
  • Privacy
Creative Commons License BY
Creative Commons Attribution 4.0 International License

Except where otherwise noted, this work is subject to a Creative Commons Attribution 4.0 International License,
which allows anyone to share and adapt our material as long as proper attribution is given.
For details and exceptions, see the Harvard Library Copyright Policy
©2024 Presidents and Fellows of Harvard College.