The Online Discovery Problem and Its Application to Lifelong Reinforcement Learning thumbnail
Pause
Mute
Subtitles not available
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

The Online Discovery Problem and Its Application to Lifelong Reinforcement Learning

Published on Jul 28, 20152538 Views

We study lifelong reinforcement learning where the agent extracts knowledge from solving a sequence of tasks to speed learning in future ones. We first formulate and study a related online discovery

Related categories

Chapter list

The Online Discovery Problem and Its Application to Lifelong Reinforcement Learning00:00
Lifelong Learning Example: Intelligent Tutoring Systems - 100:13
Lifelong Learning Example: Intelligent Tutoring Systems - 200:30
Lifelong Learning Example: Intelligent Tutoring Systems - 300:36
Lifelong Learning Example: Intelligent Tutoring Systems - 401:02
Lifelong Learning Example: Intelligent Tutoring Systems - 501:09
Lifelong Learning Example: Intelligent Tutoring Systems - 601:13
Lifelong Learning Example: Intelligent Tutoring Systems - 701:20
Task as Finite Markov Decision Process (MDP)01:34
A Class of Lifelong RL Problems - 102:06
A Class of Lifelong RL Problems - 203:27
A Class of Lifelong RL Problems - 303:46
Two Kinds of Exploration - 104:30
Two Kinds of Exploration - 205:01
Two Kinds of Exploration - 305:11
Two Kinds of Exploration - 405:23
Two Kinds of Exploration - 505:56
The Online Discovery Problem: Abstraction of Cross-task Exploration - 106:18
The Online Discovery Problem: Abstraction of Cross-task Exploration - 207:20
Explore-First Algorithm 08:52
Forced-Exploration Algorithm 10:29
A Lifelong RL Algorithm based on FE11:38
Sample Complexity of Exploration13:08
Experiment15:15
Stochastic Setting with small µm16:46
Adversarial Setting with Changing Distribution17:45
Conclusions18:33