DS677.002 - Deep Learning (SPRING · 2024)

• Description
• Teaching Team
• Logistics
• Materials
• Schedule

Description (DS677.002)

Deep Learning (DL) is a subfield of Machine Learning that has delivered disruptive technologies, and created AI algorithms that outperform humans in various tasks. It paves the way for broader advances in science. DL consists of a set of specialized techniques that exploit the abundant availability of data and computational power to build models that are composed of multiple processing layers and learn representations of data at multiple levels of abstraction. Only a few years back, the development of DL models required significant experience, but the introduction of open-source DL libraries like TensorFlow and PyTorch has opened the area to scientists and professionals with more diverse backgrounds. The course opens with a review of Artificial Neural Networks that guides you through PyTorch and enables you to build novel ANN architectures. Then it presents the evolution of progressively deeper architectures for Convolutional Neural Networks, that addressed various training difficulties and led to very successful image classification models. In this spatial context, you will also learn about Generative Adversarial Networks that are behind the fascinating “Deep Fake” images and videos. The course then takes you to the emerging applications of Recurrent Neural Networks in temporal data, including Natural Language Processing. You will also learn about Graph Neural Networks and their applications in the analysis of real-world networks (e.g., social, or biological networks). The course may also touch upon selected topics like the ability of deep networks to generalize, techniques for 'pruning' deep networks to make them more computationally efficient, and successful applications of DL methods in the Sciences.

Teaching Team

Logistics

• Time: Friday, 2:30pm - 5:20pm EST (Spring 2024）
• Location: ME 224
• Mode: Face-to-Face
• Discussions: Feel free to participate in different discussion forums on Canvas
• Quizzes & Assignments: submit and post on Canvas
• Prerequisites: CS675 or DS675 or instructor permission.
• Contact: For external inquiries, personal matters, or emergencies, please email the instructor or TA directly.

Materials

There is no required course textbook. The course will draw material from several sources, including the instructor’s own notes. Some open resources include:

• Aston Zhang, Zachary C. Lipton, Mu Li, and Alexander J. Smola, Dive into Deep Learning
• Prince, S.J. Understanding Deep Learning. MIT press. (Dec 5, 2023)

Grading

• Quizzes [10%]: Weekly Canvas quizzes reinforcing the material of each module, will help you keep up with the most important theoretical concepts. These quizzes are not proctored.
• Class Participation [10%]: You are expected to attend classes and participate in classes by listening and understanding class contents and asking related questions. You are also expected to participate in weekly Canvas discussions prompted by the instructor, with meaningful questions and answers related to the week’s topics or assignments.
• Homeworks [30%]: Assignments will be given biweekly to give you an opportunity to apply course concepts for that week. Four homework assignments of equal grading weight.
• Mid-term Exam [15%]: In-person exam, 90 minutes. Students are expected to bring a fully charged laptop, as the exam will be on Canvas with LockDown browser. Each student is allowed to bring at most 5 pages of notes. In the event the exam has to take place online, Respondus Monitor will be used for proctoring.
• Project [35%]: The project will consist of three milestones, with weights [10%, 5%, 20%]. You will have opportunities to iterate and revise your work based on peer, TA, and instructor’s feedback.

Schedule

The lecture slides are available for download in the Google Drive folder, while all weekly quizzes and biweekly assignments can be accessed on Canvas.

Week	Date	Topic	Textbook Readings	Due Work
Week 1	1/19	Introduction and course overview	chapter 2, chapter 3
Week 2	1/26	Basic Building blocks in Deep Learning (Softmax Regression, Regularization, Dropouts, Initialization, optimization, etc.)	chapters 4, 5, chapters 3, 5	hwk #1
Week 3	2/2	Convolutional Neural Networks (CNNs)	chapter 7	project initiation
Week 4	2/9	Modern CNNs	chapter 8	hwk #2
Week 5	2/16	RNNs and Language Modeling	chapter 9
Week 6	2/23	Modern RNNs and Language Translation	chapter 10	hwk #3
Week 7	3/1	Transformers	chapter 11	project milestone #1
Week 8	3/8	Midterm exam		Summary quiz, hwk #4
Week 9	3/15	Spring Break (No class)	chapter 12	project milestone #2
Week 10	3/22	Vision Transformers		hwk #5
Week 11	3/29	Good Friday (No Class)
Week 12	4/5	Graph Neural Networks	online tutorial; chapter 14	hwk #6
Week 13	4/12	Reinforcement Learning	chapter 17	project milestone #3
Week 14	4/19	Deep Generative Modeling	chapter 20
Week 15	4/26	New Frontiers in Deep Learning
Week 16	4/30	Project Presentation		Project presentation

The schedule is subject to change. Please check Canvas for the most up-to-date information.

For more detailed course information, please refer to the syllabus and the Canvas modules!