Majority-of-Three is Optimal [arXiv]
Proving the Majority-of-Three Conjecture
TL;DRs
These are (hopefully) more accessible writeups of my research, primarily aimed at readers comfortable with math and ML but not theory (or who don't want to read the entire paper).
I started writing these mostly to get better at explaining things without defaulting to heavy math; to that end, the writeups here are intended to convey just the main results and intuition. If you read one of these and want more detail, the paper is always linked.
A full list of work is at /research.
3️⃣
🐎
A Theory of Saddle Escape in Deep Nonlinear Networks [arXiv]
Analyzing training dynamics in small-initialization deep nonlinear neural networks
🔇
Rao-Blackwellized Score Matching on Manifolds [arXiv]
Figuring out what denoising score matching actually learns when data is drawn from a low-dimensional structure in a high-dimensional space