jimmYA-1995/NCTU_DeepLearning

NCTU_DeepLearning

Text Book: https://www.deeplearningbook.org

Mentioned software: python, numpy, sklearn, tensorflow, keras

google colab

Week01-Week02 Introduction and Feedforward networks

slides:

• https://drive.google.com/a/nctu.edu.tw/file/d/1VTuiNdYzNJ6KRDWKPI6B-lVUNhZwKKF1/view?usp=sharing
• https://drive.google.com/a/nctu.edu.tw/file/d/1mQgQlqh2x3MgrOCBt7XMYG4nw29upGyR/view?usp=sharing
• https://drive.google.com/file/d/1gpG9-btRo0LyLn8r8aaOozo0eOvgPMKh/view?usp=sharing

Coursework

submit form: https://docs.google.com/forms/d/e/1FAIpQLScGgaTCzsXwxhuxizLduURIBNGlOgedYCOb5gkJYwsHoAouRA/viewform

Please submit .ipynb files

• Use nearest neighbor method to do handwritten digit recognition
• (Optional) Use PCA to speed up the above method
• Handcraft a feedforward neural network that solves the problem of
  • input a binary representation of a number and classify by it's remainder when divided by 4 (with 100% accuracy)
  • input a binary representation of a number and classify by it's remainder when divided by 3 (with high accuracy)
  • input is a 3x3 board, each cell is either white or black. Check whether there are any 3 white cell are in a row (like the game tic-tac-toe)

See https://github.com/tjwei/CrashCourseML

• Basic_ML/01-From NumPy to MNIST.ipynb
• DIY_NN/01-FeedForward-Forward Propagation.ipynb

Week03 Cost function and Gradient Descent

• ipynb: https://github.com/tjwei/CrashCourseML/blob/master/DIY_NN/02-FeedForward-Backpropagation.ipynb
• Ian Goodfellow's slides: http://www.deeplearningbook.org/slides/sgd_and_cost_structure.pdf Coursework (use new E3 to submit your work)
• Use numpy to write simple neural networks and use a gradient descent algorithm to train it for classifying digits in the MNIST dataset.
  • Using cross-entropy loss
  • Using mean-square-error loss

Week04 CNN and Using tensorflow 2.0

• Train a neural network to classify digits in the MNIST dataset.
• Train a neural netowork to classify classes in the cifar10 dataset.

Week05 Convolution and convolution transpose

• Handcraft a CNN network to generate the next step of the game of life.

Week06 More Techniques

• gradient descent based optimizers: http://ruder.io/optimizing-gradient-descent/index.html
• Batch normalization: https://en.wikipedia.org/wiki/Batch_normalization https://arxiv.org/pdf/1502.03167v3.pdf
• Overfitting
• Dropout: http://jmlr.org/papers/volume15/srivastava14a/srivastava14a.pdf
• network in network: https://arxiv.org/pdf/1312.4400.pdf
• Resnet: https://arxiv.org/pdf/1512.03385.pdf

Week07 Optimization, RNN

• http://www.deeplearningbook.org/contents/optimization.html
• http://www.deeplearningbook.org/contents/rnn.html
• https://colah.github.io/posts/2015-08-Understanding-LSTMs/
• https://www.tensorflow.org/tutorials/text/text_classification_rnn

Week09 Applications

• slides: Deep Learning Week 09.pdf

Week12 Reinforcement Learning:

• Book: http://incompleteideas.net/book/bookdraft2017nov5.pdf

Week13 Deep Reinforcement Learning:

• An Introduction to Deep Reinforcement Learning: https://arxiv.org/pdf/1811.12560.pdf
• slides: https://www.slideshare.net/BigDataColombia/an-introduction-to-deep-reinforcement-learning , https://www.cse.cuhk.edu.hk/irwin.king/_media/presentations/introduction2drl.pdf
• Playing Atari with Deep Reinforcement Learning: https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf
• Double DQN: https://www.aaai.org/ocs/index.php/AAAI/AAAI16/paper/download/12389/11847
• Dueling Network Architectures: http://proceedings.mlr.press/v48/wangf16.pdf
• Ditributional DQN:
  • A Distributional Perspective on Reinforcement Learning: https://arxiv.org/abs/1707.06887
  • Distributional Reinforcement Learning with Quantile Regression: https://arxiv.org/abs/1710.10044
  • An Analysis of Categorical Distributional Reinforcement Learning: https://arxiv.org/abs/1802.08163

Week14 Generative Adversarial Networks

• https://github.com/tjwei/GAN_Tutorial
• slides: https://drive.google.com/open?id=1---Gat7n_6s0_3m_mJnaJIdRgs7XCg0U