In this course, we will study and develop techniques for accelerating large-scale video analytics using deep learning. We will study the internals of modern deep learning architectures with a focus on learning end-to-end models for video analytics.
- Instructor: Joy Arulraj
- Time: Mon/Wed 4:30 – 5:45 PM
- Location: Weber Space Science and Technology Building III (Room 1)
- On-line discussion: Piazza
- Grading tool: Gradescope
- TAs: Jaeho Bang, Gaurav Tarlok Kakkar
- Office hours:
Joy: Mon/Wed 3:30 – 4:30 PM (in Klaus 3324)
Jaeho: Tue 3:30 – 4:30 PM (in Klaus 3324)
Gaurav: Thu/Fri 3:00 - 4:00 PM (in Klaus 3324)
The course is a combination of lectures and programming projects in which we will study the internals of modern deep learning architectures with a focus on learning end-to-end models for video analytics. The students will learn to design and develop their own deep learning models and gain a broader perspective on cutting-edge research in video analytics. The course will provide research opportunities in the areas of video analytics, deep learning, and data management.
- Proficiency in Python (strict) [Tutorial]
- Basic Probability and Statistics (strict)
- Introduction to Machine Learning (recommended)
- Introduction to Database Systems (recommended)
- Calculus and Linear Algebra (recommended)
- High-level familiarity in C/C++ (recommended)
The course is open to both graduate and undergraduate students.
- Academic Honesty: Students are expected to abide by the Georgia Tech Honor Code.
This is a graduate-level course on the internals of deep learning architectures and their application to video analytics. This course has a heavy emphasis on programming projects. Upon successful completion of this course, the student should be able to:
- Understand and apply state-of-the-art implementation techniques for deep learning models following modern coding practices.
- Identify trade-offs among deep learning architectures and contrast alternatives for diverse video analytics tasks.
- Develop and justify design decisions in the context of a high-performance video analytics system.
- Implement and evaluate complex, scalable components of video analytics systems, with emphasis on providing experimental evidence for design decisions.
The final grade for the course will be tentatively based on the following weights:
- 30% Assignments
- 30% Midterm Exam
- 40% Group Project
The slide decks are derived from the following courses:
- Stanford 231N: Convolutional Neural Networks for Visual Recognition, Fei Fei Li, Andrej Karpathy, and Justin Johnson.
- CMU 15-721: Advanced Database Systems, Andy Pavlo.
This website is based on a design by Andy Pavlo.