TheoremPath is a structured way to study machine learning from axioms to frontier topics.

I built TheoremPath because I wanted a more structured way to learn machine learning.

I spend a lot of time reading ML papers, collecting ideas, organizing notes, and trying to understand how the field fits together. Over time I kept running into the same problem: there is a huge amount of material, the pace of research is intense, and new results become public constantly. You can read a lot, build a lot, and still not have a clear sense of what depends on what, where your gaps are, or what to study next.

For years I have kept detailed notes, collected textbooks, and treated technical subjects as things worth organizing carefully rather than consuming once. My background is in mathematics and statistics, and I have always learned best through structure, repetition, and working weak spots until they stop being weak.

At the same time, part of what pulled me deeper into ML was building. During graduate school I went through Andrew Ng's Deep Learning Specialization, and the advice to start small, get something working, and then add features stuck with me.

Another influence came from actuarial exam prep. I did not study most of that material through formal classes, so I had to learn it in a structured, self-directed way. That taught me the value of breaking a large subject into parts, drilling weak areas, adapting practice based on performance, and building an honest sense of readiness.

TheoremPath grew out of all of this: years of note-taking, collecting technical material, reading ML papers, organizing ideas, and trying to build better study tools.

This site does not replace hard study. Real understanding still takes writing things out, drilling, coding, debugging, and revisiting ideas properly. The point is to make that process more structured, more directed, and less random.