Lectures

You can download the lectures here (in PDF format). I will try to upload lectures prior to their corresponding classes.

• 

  Introduction to Deep Learning
  tl;dr: What is deep learning? How is it different from machine learning? What is data representation?
  

  [slides] [notes]
• 

  Mathematical Building Blocks of Neural Networks
  tl;dr: What are building blocks of networks, i.e. tensors? How to train a network with gradient descent and backpropagation?
  

  [slides] [notes] [code]
• 

  Getting Started with Neural Networks
  tl;dr: Getting a taste of deep learning with three examples: sentiment analysis, text classification, and regression.
  

  [slides] [Predicting house prices] [Sentiment Analysis: IMDB movie reviews] [Classifying Newswire]
• 

  Computer Vision
  tl;dr: A review of ConvNets, data augmentation, pre-training, fine tuning, and visualization.
  

  [slides]
• 

  Fundamentals of Machine Learning
  tl;dr: A general overview of ML, its branches, data representation, feature engineering, overfitting and regularization.
  

  [slides]
• 

  Processing Text and Sequential Data
  tl;dr: Recurrent models, SimpleRNN, LSTM, GRU, word embeddings, vanishing gradient problem, attention mechanism, and ID ConvNets
  

  [slides] [IMDB sentiment analysis using MLP with pre-trained embeddings] [IMDB sentiment analysis using MLP with SimpleRNN] [IMDB sentiment analysis using MLP with LSTM] [Temperature forecasting problem using GRUs] [Text processing using 1D ConvNets]
• 

  Advanced Models
  tl;dr: Multi input/output models, directed acyclic graphs, weight sharing, callbacks, early-stopping and checkpointing, and Tensorboard.
  

  [slides] [Keras functional API]
• 

  Generative Deep Learning
  tl;dr: Text generation with LSTMs, DeepDream, Neural Style Transfer, concept space, variational autoencoders (VAE), generative adversarial networks (GAN)
  

  [slides] [Text generation with LSTMs] [DeepDream] [MNIST with VAE] [GAN for frog generation]
• 

  Transformer and BERT
  tl;dr: The Transformer architecure for sequence encoding, multi-head attention, contextualised embeddings, and BERT
  

  [slides]