Deep_Learning

Deep Learning Class in NYCU, 2023.02~2023.06

HW1

Implement backpropagation in simple MLP with numpy.
model_name: model1, model2

python hw1.py --model_name ${model_name}

HW2

Use C++ to train 2048 AI with TD(0) and n-tuple network
Alpha: 0.1, epochs: 1 to 150K
Alpha: 0.01, epochs: 150k+1 to 250K
2048 accuracy: 0.96

g++ -std=c++11 -O3 -o 2048 2048.cpp
./2048

HW3

In BCI competition dataset use EEGNet, DeepConvNet to solve classification problem.
And using three activation layer: ReLU, Leaky ReLU, ELU to compare.
model_name: EEGNet, DeepConvNet

python train.py --model_name ${model_name}

HW4

In Diabetic Retinopathy Detection (kaggle), implement the ResNet18、ResNet50 architecture to classify imaging conditions.
model_name: ResNet18, ResNet50

python train.py --model_name ${model_name}

HW5

In bair robot pushing small dataset, implement Conditional VAE and use two past frames to predict the next ten frames.

python train.py 

HW6

Part 1: Solve LunarLander-v2 using DQN
Part 2: Solve LunarLanderContinuous-v2 using DDPG
Part 3: Solve BreakoutNoFrameskip-v4 using DQN

python dqn.py
python ddpg.py
python dqn_breakout.py