Machine Learning NTU 筆記
Introduction
機器學習基石
Week1 The Learning Problem
Week2 Learning to Answer Yes/No
Week3 Types of Learning
Week4 Feasibility of Learning
Week5 Training vs Testing
Week6 Theory of Generalization
Week7 VC Dimension
Week8 Noise and Error
Ch9 Linear Regression
機器學習技法
Ch1 Linear SVM
李宏毅 Machine Learning
Lec 4: Classification
Lec 5: Logistic Regression
Lec 7: Backpropagation
Lec 8-1: “Hello world” of deep learning
Lec 9-1: Tips for Training DNN
Lec 10: Convolutional Neural Network
Lec 11: Why Deep?
Lec 12: Semi-supervised
Lec 13: Unsupervised Learning - Linear Methods
Lec 15: Unsupervised Learning - Neighbor Embedding
Lec 17: Unsupervised Learning - Deep Generative Model (Part I)
Lec 18: Unsupervised Learning - Deep Generative Model (Part II)
Lec 19: Transfer Learning
Lec 20: Support Vector Machine
Lec 22: Ensemble
Lec 23-1: Deep Reinforcement Learning
Lec 23-2: Policy Gradient
Lec 23-3: Reinforcement Learning (including Q-learning)
2019 Life Long Learning (LLL)
2019 Meta Learning
Automatic Hyperparameter Tuning
2019 More about Auto-encoder
李宏毅 Advanced Topics in Deep Learning
Deep Learning for Language Modeling
Spatial Transformer Layer
Highway Network & Grid LSTM
Conditional Generation by RNN & Attention
Generative Adversarial Network
Improved Generative Adversarial Network
RL and GAN for Sentence Generation and Chat-bot
Ensemble of GAN
Video Generation by GAN
Ensemble of GAN
李宏毅 Deep Learning Theory
DLT 1-1: Can shallow network fit any function?
李宏毅 GAN Lecture
2017-1: Intro of GAN
2017-2: CycleGAN
2017-3: Improving Sequence Generation by GAN
2017-3: Unsupervised Conditional Generation
2017-4: From A to Z
2018-3: Unsupervised Conditional Generation
2018-7: Info GAN, VAE-GAN, BiGAN
2018-9: Sequence Generation
2018-10 Evaluation & Concluding Remarks
李宏毅 Deep Reinforcement Learning
Lec 1: Policy Gradient (Review)
Lec 2: Proximal Policy Optimization (PPO)
Lec 3: Q-learning (Basic Idea)
Lec 4: Q-learning (Advanced Tips)
Lec 5: Q-learning (Continuous Action)
Lec 6: Actor-Critic
Lec 7: Sparse Reward
Lec 8: Imitation Learning
清大吳尚鴻 Deep Learning
ML-17: Deep Reinforcement Learning
Prioritized Replay
CS231n - CNN for Visual Recognition (2017)
Lec 3 | Loss Functions and Optimization
Lec 10 | Recurrent Neural Networks
Lec 11 | Detection and Segmentation
Lec 12 | Visualizing and Understanding
Lec 16 | Adversarial Examples and Adversarial Training
李宏毅 Deep Learning Theory
Powered by
GitBook
Lec 5: Q-learning (Continuous Action)
Youtube
一般來說，Q learning 只能應付離散的值
第三招
a 是一個 continuous 的 vector，例如機器人的每個關節所需要轉的角度
Q function
什麼樣的 a 會讓 Q value 最大呢? 其實就是
只有在 必須是正定(positive definite)矩陣的情況下才成立
第四招
不要用 Q-learning
Actor-Critic 才是王道
results matching "
"
No results matching "
"
