Roadmap for Learning AI

Inspiration

Five years after graduate school, I found myself feeling stagnant, searching for a challenge that would push me beyond my comfort zone and expand my skill set. So in 2023, I decided to dive into AI. Why AI? Well, in my perspective, it struck the right balance—technically demanding, intellectually stimulating, and full of potential. But with an overwhelming sea of resources, finding a clear path wasn’t easy. So, I began documenting my journey, hoping to share it as a blueprint for others on a similar path or simply looking for a place to start.

Foundations

Programming

As a Berkeley graduate (Go Bears!), I can’t recommend CS 61A enough. This course lays a solid foundation in programming, covering fundamental concepts that are essential for anyone diving into AI and beyond. The lecture videos can be accessed here.

Calculus & Linear Algebra

Stay tuned.

Probability & Statistics

Stay tuned.

Theory

Machine Learning

For traditional ML, Stanford's CS229 by Andrew Ng is often recommended—and while it's an excellent course, I personally preferred Professor Kilian Weinberger's teaching style in Machine Learning for Intelligent Systems. Having casually reviewed several ML courses over the years, I found Weinberger's to be the most comprehensive introduction to the subject.

Deep Learning

With traditional ML under your belt, it's time to dive into neural networks. Stanford's CS231n is one of the best introductions to the topic, but the widely accessible lecture videos are from 2016—now somewhat outdated given the rapid advancements in the field. I’d recommend watching CS231n up to lecture 6 (backpropagation), which features some of the clearest explanations I’ve found, delivered by none other than Andrej Karpathy. From there, I’d suggest transitioning to Michigan's EECS 498 by Justin Johnson (one of CS231n’s main instructors) for a more updated take.

Many people find the mathematical aspects of deep learning overwhelming:

For a refresher on linear algebra and probability, I recommend the relevant sections from the foundational book Deep Learning by Ian Goodfellow et al.
For matrix calculus, The Matrix Cookbook is an excellent resource, along with its youtube walkthroughs: Part 1 & Part 2.

For supplementary information, I recommend exploring the Neural Network series by 3Blue1Brown and StatQuest. Both channels offer clear and engaging explanations that can help solidify your understanding.

Natural Language Processing

For an introduction to Natural Language Processing, I recommend starting with Stanford’s CS224N. If you’ve already taken the deep learning course, you’ll notice some overlap—particularly with RNNs, LSTMs, and Transformers, which form the backbone of modern NLP. Feel free to skip lectures covering familiar concepts. You can access the lecture videos here.

Once you have a solid foundation, the next step is to explore advanced NLP techniques, state-of-the-art (SOTA) models, and the process of building complex NLP systems. More details and resources coming soon.

Artificial Intelligence

Stay tuned.

Reinforcement Learning

Stay tuned.

Application

Data Structures & Algorithms

Mastering common algorithms and data structures is one of the most empowering aspects of a computer science education. It directly enhances problem-solving skills, and the concepts you learn will naturally extend to building real-world applications. For this, I have to go back to my Berkeley roots with CS 61B, taught by one of my favorite professors, Josh Hug.

ML Engineering

Stay tuned.

AI Engineering

Stay tuned.