This is repository for Udacity Deep Learning Nano Degree
- ENV : Anaconda + Jupyter notebook + python 3.5 + TF 1.7.0_GPU_Verion
- Using Pandas API to load and understand your data
- Understand Nerual Network like input layer, hidden layer and output layer
- Understand activate function like sigmoid, tang and reLu.
- Understand pros and cons regarding above activate function
- Understand loss function like MSE, log loss etc.
- Understand forward propagation and backward propagation
- Build your first Nerual Netword and implement forward propagation and backward propagation
- Train your Nerual Netword and play with arguments like learning rate, epoch, hidden layer nodes etc.
- Test your model and answer what do you think of your model and why
- Refer to Machine Learning Project 5(Same project).
- ENV : Anaconda + Jupyter notebook + python 3.5 + TF 1.0.0_CPU_Version (Chinese version)
- Understand recurrent nerual network
- Understand Long Short Term Memery(LSTM) cell
- Understand hyper parameters and know how to tune these parameters(learning rate, mini batch size, epoch and hidden nodes, layers etc)
- Understand and implement Mini project Word2Vec
- Understand and implement Mini project sentiment-rnn
- Implement preproccess step in P3
- Implement embed layer and LSTM layer
- Implement get batch method
- Put everything together and train your RNN
- Generate your own tv script
- ENV : Anaconda + Jupyter notebook + python 3.5 + TF 1.7.0_GPU_Verion
- Understand GAN and implement GAN in MINI project
- Understand Batch Normalization and DCGAN
- Implement Batch Normalization and DCGAN in MINI project
- Understand and Implement semi-supervised learning GAN (optional)
- Implement P4 Face generation (almost same as MINI project DCGAN, except you need to implement training process by your own)
- ENV : Anaconda + Jupyter notebook + python 3.5 + TF 1.7.0_GPU_Verion
- Understand MDP
- Understand MC and TD regarding RL
- Understand Actor-Critic
- Implement RL-Quadcopter (can use almost 90% code from examples but need design your own reward function, NN and tune some hyper parameters)