Course Description: As data becomes ubiquitous and computing resources cheaper, many disciplines have turned to deep learning to solve complicated problems. While it has achieved remarkable success in a variety of “human” tasks, deep learning is often treated as a black-box. In this course, we will study deep learning from its foundations and build an intuitive understanding for why it works. Pairing lectures with labs, we will develop cutting-edge deep learning solutions to a variety of real-world problems. We will cover neural networks, convolutional networks designed for object detection, and recurrent networks used for natural language processing. We may also explore other topics including transformers, reinforcement learning, and generative adversarial networks subject to time and interest.
Prerequisites: I will assume familiarity with linear algebra and calculus. In order to ensure we’re on the same page, please review the Wikipedia articles on matrix multiplication and gradients. We will be coding in python and I will expect you to be familiar with it.
Structure: We will meet on Monday, Tuesday, Wednesday, and Thursday at 75 Shannon in Room 202. The lecture is from 10am to 12pm and the lab is from 2 to 3pm. I will hold my office hours directly after the lab.
Resources: This class uses material from Chinmay Hegde’s phenomenal graduate deep learning class at NYU Tandon. For each class, I will post my handwritten notes, the python notebook for the demo, and the written material I used to prepare. If you would like an additional resource, I have heard the free online textbook Dive into Deep Learning is excellent.
Your grade in the class will be based on the number of points you earn.
Participation and Questions (14 points): Since winter term classes are smaller, let’s take advantage of the opportunity for more engagement. Unless you have a reasonable excuse (e.g. sickness, family emergency), I expect you to attend every lecture and demo. Whether you are able to attend or not, I expect you to fill out the form linked from the home page to receive credit for engagement (one point per day). Of course, if you are not able to attend in person, you should read the reading and run the demo.
Homeworks (56 points): There will be one homework problem per class. In order to encourage engagement with the solutions, your homework grade will be based both on your solutions and your self-grade of your solutions.
Part 1: Solutions (3 points per homework problem)
You will submit your solutions to the homework problems via Canvas. Writing clean and understandable math is an important skill so a component of your homework grade is to typeset your solutions in LaTeX. The homework (and LaTeX source) is posted on the homepage. I suggest working in Overleaf to avoid the hassle of setting up the requisite LaTeX software on your own computer.
Part 2: Self-grade (1 points per homework problem)
In order to encourage engagement with the solutions, you will reflect on what you did well in the homework and what you learned. You should submit the self-grade after comparing your answers to the solutions.
Project (30 points): An exciting part of deep learning is its potential to solve complicated and “human” tasks. You will brainstorm and execute a project on a data set of your choice. Your project should be related to an architecture or task that we have covered in class but you will be expected to take it to the next level. You can complete your project as an individual or with a partner.
Late Policy: You have one day (24 hours) of no-questions-asked total “lateness.” I will total your lateness at the end of the class and each day of lateness beyond your allowance will result in a 5 point drop in your grade.
Academic integrity is an important part of your learning experience. You are welcome to use online material and discuss problems with others but you must explicitly acknowledge the outside resources on the work you submit.
If I notice that you have copied someone else’s work without proper attribution (such as code from the internet without a reference link or a solution very close to another student’s without giving credit), I will give you a warning. After the warning, I will subtract 5 points for every violation.
If you have a Letter of Accommodation, it is your responsibility to contact me as early in the term as possible. If you do not have a Letter of Accommodation and you believe you are eligible, please reach out to the ADA Coordinators at ada@middlebury.edu.