I am starting this space to collect ideas that are useful before they become polished papers: questions from ongoing projects, implementation details that are easy to forget, and connections between different areas of machine learning.
My current interests center on generative models, world modeling, and multimodal learning. I am especially curious about how scientific structure and prior knowledge can help these models learn more reliably and generalize beyond familiar data.
Posts here will be concise and practical. Some will explain a paper or concept; others will document an experiment, a useful tool, or an open question. I hope these notes can make my research process more legible—and occasionally be useful to someone working on a related problem.
More soon.