In recent years, deep learning has enabled huge progress in many domains. The aim of this lecture is to introduce the fundamentals of current deep learning. In this lecture, students will learn the visual recognition using Convolutional Neural Network (CNN), the sequence models such as Recurrent Neural Networks(RNN), the generative models such as Generative Adversarial Network(GAN) and Deep Reinforcement Learning (DRN). Students will also learn the practical programming in Tensorflow and Keras. During the course, students will do three team projects such as CNN, RNN, and GAN by forming project teams.
- Training neural networks, Keras and Tensorflow
- Visual recognition: CNN (convolution, padding, stride, pooling), LeNet, Alexnet, Resnets, Inception network, Transfer learning, Data augmentation, Object localization, Landmark detection, Sliding windows, YOLO, One shot learning, Siamese network, Face verification, Cost functions, Autoencoders, Region proposal, Segmentation
- Sequence models: Sentence classification using CNN, RNN, LSTM(Long Short-Term Memory Networks), Embeddings and Word2Vec, Attention model, Image Caption Generation using LSTM, Neural Machine Translation
- GAN: GAN model, Deep convolutional GAN, Conditional GAN, Neural Style transfer, WaveNet
- Deep Reinforcement Learning: Framework, Dynamic programming, Monte carlo method, Temporal Difference Learning, SARSA, Q-learning, Deep Q-learning, Actor-critic