CS7545_Sp24_Lecture_01 - mltheory/CS7545 GitHub Wiki

CS 7545: Machine Learning Theory -- Spring 2024

Instructor: Jacob Abernethy

Notes for Lecture 01

January 09, 2024

Scribes : None

Professor got sick right before class!

We had to cancel. Instead, the post below was shared on Piazza.

Welcome students to CS 7545!

First, my apologies for missing my first day of class. It was really a bad time to get that little pink line on the covid test.

For Thursday's lecture, I expect we'll do virtual -- please keep your eye out for a post on this topic tomorrow. 

The good news is that the first day of class is usually just me reciting high-level descriptions about the class, basically what you can already read on the course website. So, really, before finishing this just go read the course page. (Even if you've seen it, I just added a bunch of new content!)

OK you read the course page carefully? Great! Here's the additional stuff I emphasize in lecture:

  • Machine Learning Theory is about trying to understand "why do the algorithms work?" and "can the success of ML be mathematically justified?" We actually have a few decades of research on this topic, and in some circumstances we really have good answers. For many of the most popular methods in use today, we have far fewer answers, and honestly there's a lot of research trying to understand what's going on. In this course, we'll give you a sample of some of the coolest ideas in ML Theory, and try to lay a foundation for you to engage in future research questions.

  • This is not an easy course and requires serious mathematical training. Over the past 5-10 years there has been huge growth in interest in any course with "machine learning" in the title. I just want to warn students that 7545 is really more of a graduate math class. While we don't expect you to have all the prerequisites for the course, we expect a certain level of mathematical sophistication.

  • How can you tell if you're ready for the course? I think the best way to judge is to look at a previous problem set from a previous year. The last time we taught this course, this homework set was due in the first 2-3 weeks of the course. If you would struggle to even begin with those problems, I might think twice about jumping in.

  • Are you on the waitlist and hoping for some kind of magic trick to get registered? Here's the trick for you: wait for Friday afternoon. I've taught this course now 5 times, and every time I get 25 emails asking to get off the waitlist, and every time I do nothing and the waitlist clears and there are open slots on Friday afternoon. So please relax, and wait for Friday.

  • How hard are the exams gonna be? When we make the exams, we try to come up with modified versions of the problems given out in lecture, including the homework problems. As long as you're following in the material in class, and you feel comfortable with the types of problems we give out, you shouldn't have any trouble.

  • Rapid fire round:

    • Will you post office hours? Yes soon.
    • What if I can't make exam X in person? We can approve certain exceptions to the in-person policy.
    • Do I have to attend every lecture? Encouraged, but not required. Scribe notes will be available for each lecture.
    • How should I communicate with you or other course staff? No email please! Piazza for general questions, or private all-staff questions. MS Teams message for more personal requests. 

OK that's all I have for now. Still have questions? Please feel free to add them to the bottom of this post and we'll do our best to answer them.

Looking forward to a fun semester!

⚠️ **GitHub.com Fallback** ⚠️