(해당 포스트는 Coursera의 Prediction and Control with Function Approximation의 강의 요약본입니다) 이전 포스트들을 통해서 설명하고자 했던 것은 기존의 Q-table과 같은 Tabular 방식이 아닌, Value를 하나의 Function, 즉 Value Function으로 근사하는 방법이 존재하고, 이때 이 근사된 Value Function과 실제 Value Function과의 오차를 줄일 수 있는 방법으로 Gradient Descent를 적용할 수 있다는 것이었다. 그래서 Function Approximation을 Monte Carlo에 적용한 Gradient MC과 제한적이기는 하나, Gradient를 TD Learning에 적용한 Semi-Gradien..
(해당 포스트는 Coursera의 Prediction and Control with Function Approximation의 강의 요약본입니다) 역시 Function Approximation 기법을 Monte Carlo Method처럼 TD Learning에다가도 접목시킬 수 있다. 우선 Monte Carlo method에 Function Approximation을 접목한 Gradient Monte Carlo에서 weight이 update되는 과정을 다시 돌아보면 다음과 같다. $$ \mathbf{w} \leftarrow \mathbf{w} + \alpha[ G_t - \hat{v}(S_t, \mathbf{w})] \nabla \hat{v}(S_t, \mathbf{w}) $$ 이를 사용하면 estima..
- Total
- Today
- Yesterday
- reward
- Kinect
- SketchFlow
- PowerPoint
- bias
- RL
- End-To-End
- arduino
- Offline RL
- ColorStream
- Gan
- Expression Blend 4
- Pipeline
- Windows Phone 7
- Off-policy
- TensorFlow Lite
- Python
- DepthStream
- Kinect SDK
- processing
- 파이썬
- 딥러닝
- Policy Gradient
- Variance
- windows 8
- 강화학습
- 한빛미디어
- dynamic programming
- Kinect for windows
- Distribution
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | |||||
3 | 4 | 5 | 6 | 7 | 8 | 9 |
10 | 11 | 12 | 13 | 14 | 15 | 16 |
17 | 18 | 19 | 20 | 21 | 22 | 23 |
24 | 25 | 26 | 27 | 28 | 29 | 30 |