Part 1
1) Course Welcome and Instructor Introductions
My name is Tony Ma. This quarter, the course will be taught by two instructors: me and Chris. I work on machine learning and machine learning theory, including theory across different areas of machine learning—reinforcement learning, representation learning, supervised learning, and related topics.
I’d like Chris to introduce himself as well.
Chris: I’m Chris, also in the machine learning group. I’m especially interested in how the systems we build are changing because of machine learning. It has been a fascinating time over the last 10 years. I started out thinking a lot about optimization and how we scale up large models, back when machine learning had relatively few applications in everyday life. Over the last couple of years, we’ve built things that, hopefully, some of you have used. My students have contributed to products like Search, Gmail, Assistants, and other areas. More recently, I’ve been focused on how to make these models robust. We’ll also have a great new lecture from Tengyu on what are called foundation models—large, self-supervised models that have been getting a lot of attention.
Percy, Tatsu, and I co-taught a course about foundation models last term. This course is exciting because it gives you the foundational layer of machine learning that all of those developments are built on. It’s a great time to study machine learning because it’s no longer abstract: you use machine learning products every day. The goal is that you’ll gain insight into how they work and why there is still so much research left to do. I’m really looking forward to lecturing you all.
2) How the Teaching Will Alternate
You’ll see Chris and me alternate every few weeks. Next lecture will be Chris, and then after two or three weeks you’ll see me again.
3) Teaching Team Structure
Let me introduce the teaching team. We will have 12 TAs, plus one head TA and a course coordinator. The head TA will also serve as the course coordinator. They will handle much of the behind-the-scenes work, so you may not need to interact with them often, but they will be organizing the whole TA team.
We currently have 12 TAs, and we may add more if enrollment increases. I didn’t ask the TAs to attend the first lecture, partly because they would also need to wear masks, and the pictures on the slide serve a similar purpose. You will see them often in office hours and other course settings.
4) What Today’s Lecture Covers
In this lecture, I’ll spend the first part on logistics and the overall course structure. Then I’ll introduce, at a high level, the topics covered in the course.
5) Where to Find Course Information Online
We worked hard to make the course information available online in a single place. The course website links to several Google Docs: one for logistics, another for the syllabus, and also links to lecture notes and final project guidelines. In principle, everything I’m presenting today is a subset of what’s on the website—and it’s a small subset. I encourage you to read through those documents, especially when you have questions. A good first step is to check whether the documents already answer what you’re wondering about.
After that, feel free to ask questions.
6) Prerequisites and Background Expectations
The first topic is prerequisites. You’ll hear different opinions about difficulty: some students say the course is challenging, others say it’s easier, often depending on their background. That is why I want to emphasize prerequisites early. Having the right background will help you meet your goals in the course.
Probability: The most important prerequisite is probability at the level of CS109 or Stats. You should at least have seen terms like dispersion, random variables, expectation, conditional probability, variance, and density. You do not need to recall everything instantly, but you should recognize these concepts from a previous course.
Linear algebra: You also need linear algebra, including matrix multiplication and eigenvectors. Linear algebra is covered in courses like Math 104, 113, and 205. The logistics document lists additional relevant courses. The key skills we rely on most are matrix multiplication and eigenvectors.
Programming: We require basic programming knowledge, especially in Python and NumPy. If you know Python but not NumPy, that is usually fine—NumPy is mostly basic numerical operations. If you do not know Python or NumPy but you know another language like C++, you can likely transition fairly easily since a lot of the work is adapting syntax. If you have no programming background, the course will be difficult, because homeworks typically have both a math component and a programming component.
One of the most challenging situations on homeworks is when your code goes wrong—which happens all the time, even for instructors—and you’re not sure whether the issue is your math or your code. Sometimes you might think you derived the wrong equations, when the real problem is that you used NumPy incorrectly. We will review Python and NumPy in some TA lectures to refresh your skills, or to help you learn the basics, but you should come in with some programming foundation.
7) TA Review Lectures and Course Intensity
We will provide materials and TA lectures to review backgrounds. There will be three lectures each on programming, linear algebra, and probability.
This is a mathematically intense course for many students, depending on background, so consider this a heads-up. It helps if you know at least two out of the three areas (programming, linear algebra, probability) relatively well. That way you are less likely to get stuck on “entangled” issues where it is unclear whether you are missing the math or the implementation.
That challenge is also part of what makes the course exciting and rewarding.
8) Course Goal and What “Foundations” Means
The goal of the course is to give you the foundations of machine learning. This is the foundational layer, and it is also an introductory course. You do not need to have taken a machine learning course before taking this one.
At the same time, we hope that after completing the course, you will feel comfortable enough with the basics to apply machine learning to applications. If you want to become an expert in specific areas like NLP or vision, you’ll likely need additional courses, but this course is designed to set up the machine learning foundations that support broader AI and AI-related applications.
Because of that, the course covers a diverse set of topics and does involve mathematics. We will do very little in the way of formal proofs, but we will do many mathematical derivations. You will do derivations in homeworks, and we will also do derivations in lecture.
If you have questions during lecture, feel free to stop me. Also, yes—lectures are recorded, and you can find recordings on Canvas.
9) Academic Integrity and Honor Code Expectations
The second important thing I want to discuss is the honor code. It may feel awkward to mention this so early, but in the past there have been honor code violations, and it is genuinely unfortunate. It’s difficult and sad to have to report students, and I don’t want to see that happen again. That is why I want to be direct about expectations from the beginning.
If you are not intentionally violating the honor code, you generally do not need to worry. Still, here are key points (and these are also on the course website in more detail):
We encourage study groups. You may discuss homework problems with others, but you must write your solutions independently, and you must list the names of the people you discussed the homework with.
It is an honor code violation to copy, refer to, or look at written or coding solutions from a previous year. This includes (but is not limited to) official solutions from previous years, solutions posted online, solutions you or someone else wrote in previous years, or solutions for related problems. If you apply common sense and avoid intentionally doing anything inappropriate, you should be fine.
We do check code using software tools, and the course staff and TAs handle academic integrity issues when they occur. I’m not trying to stress you out, but I do want to put the policy up front.
10) Course Components: Project, Homework, Midterm, and Support Sections
Beyond homework, another major component is the course project. We encourage you to form groups of one to three people. The evaluation criteria are the same whether you work alone or in a group of two or three. More information is on the course website.
Typically, you’ll apply machine learning to an application or topic you are interested in. This is one of my favorite parts of the course. Each quarter we receive around 100 project submissions, covering a wide variety of topics and applications of machine learning. You are welcome to pursue other topics as well. You can also focus on core algorithms, which is also great, but many students explore applications in areas like music and finance.
The course also has four homeworks and a midterm. There is no final exam. The main graded components are the midterm, the course project, and the homeworks.
Another component is TA lectures, which are optional. If you find them useful, you should attend; if not, you don’t have to.
There are two sets of TA-led sessions:
Friday TA lectures (Friday section): We will likely have six to seven weeks of these. The first three weeks review foundational material, especially concepts related to machine learning. Later weeks cover more advanced topics that are not required but may be interesting.
Discussion sections: These are meant to be interactive. Since the course is large, it can be harder to make lectures interactive for every person. Discussion sections are smaller sessions led by TAs, designed to feel more like a traditional classroom. They help bridge the gap between lectures and homework. TAs will work through problems similar to homework (sometimes simpler), and the sessions will often involve live problem-solving, students presenting solutions, and discussion among students. You can find the exact time and format in the logistics Google Doc.
11) Platforms, Logistics, and Deadline Policies
There is more detailed information on the course website and the logistics Google Doc, which is quite comprehensive. Recordings are on Canvas. Canvas also has a course calendar. The syllabus page links to lecture notes.
For Q&A, we will use Ed. We strongly encourage you to use Ed for communicating with us in almost all situations. You can make private posts or anonymous posts depending on what you need. If you do not have access to Ed, you may need to email us; you can email the head TA to get access.
Homework submission will be through Gradescope. The logistics document also explains late-day policies. One important heads-up: we do not allow late days for the final project. The main reason is that, especially in spring quarter, the grading deadline is very tight—only a few days after finals week—and some students need final grades quickly due to graduation requirements. We also do not want to make the project deadline too early, because it would conflict with homework deadlines.
The final project deadline is on Monday of finals week, though you should double-check the exact date in the documents. We try to set it as late as possible while still meeting grading constraints, which is why late days aren’t allowed for the project. Additional FAQs are also in the Google Doc.
12) Student Question About Discussion Sections
Student: For discussion, will we be assigned to a specific session, or do we get to choose which discussion sessions we go to?
Answer: Currently, we have two TAs offering two discussion sessions. We will try to keep the materials in both sessions essentially the same. The times are not fully set yet, but you should be able to choose whichever session you want. It’s probably best to consistently attend one session so the TA gets to know you, but you don’t have to. Discussion sections are optional, and you should attend based on your needs.
13) Transition to the Scientific Content of the Course
If there are no other questions, we’ll move on to the more scientific part of the course. As I said, the main goal is to build your foundations in machine learning. We will cover a diverse set of topics, and we will approach them in a mathematical way.
