Blog
Notes on mathematics and machine learning.
Time-reversal refocusing in random media
Simulating wave refocusing and studying how mirror aperture and random fluctuations change the returned field.
Learning pendulum swing-up from demonstration
Fitting a dynamics model, adjusting a demonstrated motion, and handing control to a balancing controller.
When Symmetries Undermine Our Confidence in Generalization Bounds
A story of how rescaling symmetries in ReLU networks undermine PAC-Bayes generalization bounds, and our theoretical solution to restore meaningful guarantees.