The focus of this module is to introduce the concepts of machine learning with as little mathematics as possible. We will introduce basic concepts in machine learning, including logistic regression, a simple but widely employed machine learning (ML) method. Also covered is multilayered perceptron (MLP), a fundamental neural network. The concept of deep learning is discussed, and also related to simpler models.

In this module we will be discussing the mathematical basis of learning deep networks. We’ll first work through how we define the issue of learning deep networks as a minimization problem of a mathematical function. After defining our mathematical goal, we will introduce validation methods to estimate real-world performance of the learned deep networks. We will then discuss how gradient descent, a classical technique in optimization, can be used to achieve this mathematical goal. Finally, we will discuss both why and how stochastic gradient descent is used in practice to learn deep networks.

This week will cover model training, as well as transfer learning and fine-tuning. In addition to learning the fundamentals of a CNN and how it is applied, careful discussion is provided on the intuition of the CNN, with the goal of providing a conceptual understanding.

This week will cover the application of neural networks to natural language processing (NLP), from simple neural models to the more complex. The fundamental concept of word embeddings is discussed, as well as how such methods are employed within model learning and usage for several NLP applications. A wide range of neural NLP models are also discussed, including recurrent neural networks, and specifically long short-term memory (LSTM) models.

This week we'll cover an Introduction to the Transformer Network, a deep machine learning model designed to be more flexible and robust than Recurrent Neural Network (RNN). We'll start by reviewing several machine learning building blocks of a Transformer Network: the Inner products of word vectors, attention mechanisms, and sequence-to-sequence encoders and decoders. Then, we'll put all of these components together to explore the complete Transformer Network.

This week will cover Reinforcement Learning, a fundamental concept in machine learning that is concerned with taking suitable actions to maximize rewards in a particular situation. After learning the initial steps of Reinforcement Learning, we'll move to Q Learning, as well as Deep Q Learning. We'll discuss the difference between the concepts of Exploration and Exploitation and why they are important.