CS677-004: Deep Learning 

Spring 2019

Instructor: Dr. Grace Guiling Wang

TA: Xiaoyuan Liang 

Book: 

book

Authors: Ian Goodfellow, Yoshua Bengio, and Aaron Courville
Title: Deep Learning
ISBN-10: 0262035618
ISBN-13: 978-0262035613.
Publisher: The MIT Press.  

 

Course description

This course studies the architecture and algorithms of deep learning. Topics to be covered include: deep feedforward networks, deep model training optimizations, convolutional networks, recurrent and recursive nets, and deep reinforcement learning. Upon successful completion of the course, students are expected to gain a deep understanding of the fundamental concepts and principles of designing and implementing deep learning networks.

Topics

Topic Reading

Math Review

Chap 2-4

Machine Learning Review

Chap 5

Introduction to Deep Learning Chap1

Deep Forwarding Networks

Chap 6

Regularization for Deep Learning

Chap 7

Optimization for Training Deep Models

Chap 8

Convolutional Networks, Capsule

Chap 9, Paper

Sequencing Modeling (RNN, LSTM)

Chap 10

Practical Methodology

Chap 11

Reinforcement Learning

Paper

GAN

NIPS tutorial, paper

 

Group Presentation Topics (with tentative schedule)

 

Programming Assignments I

Transportation mode classification based on smart phone accelerometer data using Machine Learning Algorithms. (Data and detailed description)

 

Programming Assignments II

(Following up assignment I) Transportation mode classification based on smart phone accelerometer data using convolutional neural networks and any kinds of regularization or optimization techniques at your discretion for better performance.  (Data and detailed description)

 

Programming Project

Prediction of financial products. We use SPY, TLT, and GLD to represent equity market, bond market, and commodity markets, respectively. Students can use any neutral networks and any optimization/regularization techniques covered in the class for better prediction result.  Project is due on May 07 and we will use real market data from 05/07 to 05/16 for testing of the accuracy. (Data and detailed description)