wang-sj16 / MLDS2018SPRING

Machine Learning and having it Deep and Structured (MLDS) in 2018 spring

MLDS2018SPRING

This course has four homeworks, group by group. The four homeworks are as follows:

1. Deep Learning Theory
2. Sequence-to-sequence Model
3. Deep Generative Model
4. Deep Reinforcement Learning

Table of Contents

1. Deep Learning Theory
   • Deep vs Shallow
   • Optimization
   • Generalization
2. Sequence-to-sequence Model
   • Video caption generation
   • Chat-bot
3. Deep Generative Model
   • Image Generation
   • Text-to-Image Generation
   • Style Transfer
4. Deep Reinforcement Learning
   • Policy Gradient
   • Deep Q Learning
   • Actor-Critic

Results of Four Homeworks

1. Deep Learning Theory

• Report.pdf

1.1 Deep vs Shallow

• README

1.2 Optimization

• README

1.3 Generalization

• README

2. Sequence-to-sequence Model

2.1 Video caption generation

• BLEU@1 = 0.7204
• README
• hw2_1/report.pdf

2.2 Chat-bot

• Perplexity = 11.83, Correlation Score = 0.53626
• README
• hw2_2/report.pdf

3. Deep Generative Model

• report.pdf

3.1 Image Generation

• README
• Image Generation: 100% (25/25) Pass Baseline

./gan-baseline/baseline_result_gan.png

3.2 Text-to-Image Generation

• README
• Text-to-Image Generation: 100% (25/25) Pass Baseline

Testing Tags	./gan-baseline/baseline_result_cgan.png
blue hair blue eyes blue hair green eyes blue hair red eyes green hair blue eyes green hair red eyes

3.3 Style Transfer

• README

4. Deep Reinforcement Learning

• report.pdf

4.1 Policy Gradient

• README
• Policy Gradient: Mean Rewards in 30 Episodes = 16.466666666666665

4.2 Deep Q Learning

• README
• Deep Q Learning: Mean Rewards in 100 Episodes = 73.16

4.3 Actor-Critic

• README