By domain · 按方向
Machine Learning机器学习
Models, training, generative methods and the intuition behind them — 3 notes, primary or secondary.
- Note2026.09Action Tokenization: Discretizing Continuous Actions
The chain "continuous action → discrete token → continuous action", unpacked into the background each link requires — π₀-FAST's discrete autoregressive route against FSQ, Binning and flow matching.
- Note2026.08Autoregressive Models: BERT / GPT and Prefix-LM
Which sequence-modelling paradigms exist, what separates BERT from GPT, and why robot VLAs such as π₀-FAST use a hybrid of the two called Prefix-LM.
- Note2026.08Flow Matching ODE Solvers, Illustrated
Sampling an image is numerically integrating an ODE. Five classic solvers — Euler, Midpoint, Heun, RK4, Dopri5 — taken apart, with real integrations behind every animation.