Blog
Occasional research notes. Equations, code, and honest accounts of what didn't work. Subscribe via RSS.
-
Why a router network reduces inter-class confusion in GI imaging
The intuition (and a little math) behind coarse-to-fine routing: confusion concentrates within groups, so that's where the capacity should go.
-
Lessons from building a 9,010-video sign language dataset
What actually consumed the time — and what I'd do differently — collecting a 530-class continuous BdSL video dataset.
-
From mechatronics to computer vision: what transfers
Kinematics, control loops, and sensor debugging turn out to be surprisingly good preparation for ML research.