Self-Improving Reactive Agents Based On Reinforcement Learning, Planning and Teaching

Overview

Introduction

Reinforcement Learning Frameworks

Reinforcement learning frameworks

AHC-learning: Framework AHCON

AHC-Learning: Framework AHCON

Q-Learning: Framework QCON

Q-Learning: Framework QCON

Experience Replay

Action Models

Framework AHCON-M

Framework QCON-M

Teaching: Frameworks AHCON-T and QCON-T

The test environment

The Learning Agents

Input Representation

Output Representation

Action Models

Prevention of over-training

Experimental Results (Global Representation)

Experimental Results (Local Representation)

QCON-T results

Discussion

Limitations

Conclusions