Control tasks and optimal outputs can be reinforced by means of reward and punishment. Reinforcement agents interact with the environment or system they are supposed to control. The agent takes meaningful actions which can change the system state and receive feedback. Desired behavior results in rewarding reinforcement signal; the gent learns what to do step by step.
Fields & Methodologies
- Dynamic Programming vs. Monte Carlo
- Q-Learning
- Temporal Differencing
- RL Agents with Fuzzy Reinforcement
- Opposition-Based RL Agents
Applications
- Learning Computer Vision Tasks
- Bandwidth Allocation in Internet
- RL Agents for Control of Industrial Systems