Applied Deep Learning (2019 Spring) @ NTU
This course is lectured by Yun-Nung (Vivian) Chen and has four homeworks. The four homeworks are as follows:
- Dialogue Modeling
- Contextual Embeddings
- Deep Reinforcement Learning
- Conditional Generative Adversarial Nets
Browse this course website for more details.
- Dialogue Modeling
- Data Preprocessing
- Training and Prediction
- Results (Recall@10)
- Sequence Classification with Contextual Embeddings
- Part 1. Train an ELMo to beat the simple baseline
- Part 2. Beat the strong baseline with nearly no limitation
- Deep Reinforcement Learning
- Policy Gradient
- Deep Q-Learning (DQN)
- Actor-Critic
- Conditional Generative Adversarial Nets
- Cartoon Set
- Evaluation
- Train Condiction GANs
- Training Tips for Improvement
- Evaluate Condiction GANs
- FID Scores
- Training Progress
- Loss and Accuracy
- Human Evaluation Results
Results of Four Homeworks
3. Deep Reinforcement Learning
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/pg-ppo/openaigym.video.0.13592.video000000.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/pg-ppo/openaigym.video.0.13592.video000001.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/pg-ppo/openaigym.video.0.13592.video000008.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/pg-ppo/openaigym.video.0.13592.video000027.gif?raw=true)
3.2. Deep Q-Learning (DQN)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/dqn/openaigym.video.0.19364.video000000.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/dqn/openaigym.video.0.19364.video000001.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/dqn/openaigym.video.0.19364.video000008.gif?raw=true)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/dqn/openaigym.video.0.19364.video000027.gif?raw=true)
World\Stage |
1 |
2 |
3 |
4 |
1 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-1-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-1-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-1-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-1-4-v0.gif?raw=true) |
2 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-2-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-2-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-2-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-2-4-v0.gif?raw=true) |
3 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-3-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-3-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-3-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-3-4-v0.gif?raw=true) |
4 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-4-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-4-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-4-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-4-4-v0.gif?raw=true) |
5 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-5-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-5-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-5-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-5-4-v0.gif?raw=true) |
6 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-6-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-6-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-6-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-6-4-v0.gif?raw=true) |
7 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-7-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-7-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-7-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-7-4-v0.gif?raw=true) |
8 |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-8-1-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-8-2-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-8-3-v0.gif?raw=true) |
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw3/results/a2c-all-all/SuperMarioBros-8-4-v0.gif?raw=true) |
4. Conditional Generative Adversarial Nets
- Resnet-based ACGAN with BCE loss (resnet_1000)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw4/eval_images/resnet_1000/resnet_1000.gif?raw=true)
4.2. Human Evaluation Results
- Resnet-based ACGAN with BCE loss (resnet_1000)
![](https://github.com/JasonYao81000/ADL2019/raw/master/hw4/eval_images/resnet_1000/results2.png?raw=true)