IST 597: Foundations of Deep Learning
and Place: 2:30-5:30, Wednesday, Earth and Eng Sciences 119
Hours: Giles, 2 PM Thursdays, Westgate E350 or by request. (note change in time!)
Mali, 10-12 AM, Tuesdays, Zoom.
This is a three hour course hands on course for graduate students
that meets once a week. Students should be familiar with the
foundations of deep learning: linear algebra, numerical computation,
probability and information theory. There will be a brief introduction
to machine learning basics and computation complexity. We will cover
deep networks and their applications. We will investigate deep
feedforward networks, convolutional networks, and recurrent networks
as well as practical methodologies, including model regularization
and parameter optimization. We will also cover transformers and
adversarial models. Practical applications will be utilized
for course projects. Students will learn deep learning software from
the projects and will give presentations on chosen relevant papers.
This course is intended to prepare students to understand, design,
develop and use deep learning methods.
Students should be familiar with the basic mathematics of deep learning
(syllabus): This schedule is
subject to change. Please check it on a regular basis for
- The reading list and student team presentation order is on the syllabus.
- Most classes will have
online handouts such as powerpoint or PDF and examples of related code. It is the student's
responsibility to download that material.
Materials and References: Course materials can
be found here. There will also be links on the schedule.
Project proposal and
Final project presentation and paper: 40%
Paper reviews and
There will be 5 exercises, each varying in points, all in
Python, PyTorch or TensorFlow.
All exercises are due by midnight of the due date and must be
submitted as a zip or gzip which includes PDF and PythonScript[.py].
Late exercises will receive 50% credit for the first 24 hours and no
credit after that time. All exercise solutions must be turned in, even if late.
Failure to do so can result in a deferred grade.
Texts and Readings:
The reading list is on the schedule above. We will use
chapters and sections from the online text Deep Learning
and related papers listed in the syllabus.
- Deep Learning,
Ian Goodfellow, Yoshua Bengio, Aaron Courville, MIT Press,
- Papers where appropriate.
- Other Deep Learning books
- Books that discuss some of the advantages and disadvantages of Deep Learning are:
- For TensorFlow, there are many books and online
tutorials. Some recommended books are:
Student teams will have a research project, either proposed or assigned by Giles and Mali.
All proposed projects will have to be approved.
There will be a topic presentation, updates, a final presentation,
and a project paper in PDF and hard copy in the ACM
There will be at two detailed presentations for the project and brief updates.
For the assigned papers, each team will make a review presentation
on their assigned paper(s).
Email: All email to the instructor and TA about this
class should contain "IST597" in the subject line. For
example, the subject line might read "IST597: Question about
....". Email without this information might be deleted by spam
filters or placed in a folder to be read at a later date.
Email with the appropriate identifier will usually be read within 24
hours of being received.
Reuse: Materials from
this course can be publicly reused in other courses. This course
Educational Resources (OER).