MLDS-Note

关于Machine Learning and having it deep and structured (2018,Spring)课程的学习笔记

hw1	hw3
hw2	hw4

目录

chapter 1: Why Deep Structure

• 1-1 Can shallow network fit any function
• 1-2 Potential of Deep
• 1-3 Is Deep better than Shallow

chapter 2: Optimization

• 2-1 When Gradient is Zero
• 2-2 Deep Linear Network
• 2-3 Does Deep Network have Local Minima
• 2-4 Geometry of Loss Surfaces (Conjecture)
• 2-5 Geometry of Loss Surfaces (Empirical)

chapter 3: Generalization

• 3-1 Capability of Generalization
• 3-2 Indicator of Generalization

chapter 4: Computational Graph（coming soon）

chapter 5: Special Network Structure

• 5-1 RNN with Gated Mechanism
• 5-2 Sequence Generation
• 5-3 Conditional Sequence Generation
• 5-4 Tips for Generation
• 5-5 Pointer Network
• 5-6 Recursive Structure
• 5-7 Attention-based Model

chapter 6: Special Training Technology

• 6-1 Tips for Training Deep Network
• 6-2 Automatically Determining Hyperparameters

chapter 7: Generative Adversarial Network (GAN)

• 7-1 Introduction of Generative Adversarial Network (GAN)
• 7-2 Conditional Generation by GAN
• 7-3 Unsupervised Conditional Generation
• 7-4 Theory behind GAN
• 7-5 fGAN：General Framework of GAN
• 7-6 Tips for Improving GAN
• 7-7 Feature Extraction
• 7-8 Intelligent Photo Editing
• 7-9 Application to Sequence Generation（coming soon）
• 7-10 Evaluation

chapter 8: Deep Reinforcement Learning

• 8-1 Introduction of Reinforcement Learning
• 8-2 Policy-based Approach (Learning an Actor)
• 8-3 Proximal Policy Optimization (PPO)
• 8-4 Q-Learning (1)
• 8-5 Q-Learning (2)
• 8-6 Actor-Critic
• 8-7 Sparse Reward
• 8-8 Imitation Learning

Appendix: homework

• hw1：finish
• hw2：TODO（not recently）
• hw3：coming soon
• hw4：coming soon

说明

1. 上述内容均来自于MLDS课程，欢迎交流与讨论（可能里面会有我个人理解错误的地方，欢迎指出）
2. 可以clone到本地，再用其他markdown阅读工具（个人采用typora编辑）
3. 请勿用于其他商业用途