Overview

Description

CALENDAR DESCRIPTION:

This course aims to give students experience to emerging important areas of computing science.

COURSE DETAILS:

Machine learning has become the main framework for building programs that perform intelligent tasks. In fields such as computer vision and natural language processing, many recent successes have been achieved using neural nets with several layers, known as deep neural nets. This course is an introduction to deep neural nets, techniques for training them from data, and significant applications. While general machine learning is not a prerequisite, the course will be difficult for students without sufficient preparation. The main learning outcomes are 1) for students to have sufficient practical experience with deep learning to apply current techniques to real-life problems (2) for students to have sufficient theoretical understand of deep neural nets to analyze and improve their performance.

COURSE-LEVEL EDUCATIONAL GOALS:

• Training feedforward neural nets (backpropagation)
• Advanced training topics, including: dropout, batch normalizations, step size adapation, hyperparame
• Common architectures and their applications: convolutional neural networks, recurrent neural network
• Embeddings (skip-gram models, graph neural networks)
• Generative models: generative adversarial models, variational auto-encoders
• Comparison of neural networks with other machine learning approaches (linear classifiers, kernel met
• Adversarial attacks against neural networks
• Interpreting neural networks

Grading

Materials

MATERIALS + SUPPLIES:

• Introduction to Deep Learning, Eugene Charniak, MIT Press, 2018, 9780262039512
• Deep Learning, Goodfellow, Bengio, and Courville, MIT Press, 2016, 9780262035613, Available on-line at http://www.deeplearningbook.org

REQUIRED READING:

RECOMMENDED READING: