- kernel
- CI
- learning-theory
- research
- basics
•
•
•
•
Why testing conditional independence is so hard?
Conditional independence testing powers causal discovery and fairness analysis. But it has a strange property-some dependent and independent distributions are indistinguishable from finite data. And modern CI tests can even fabricate dependence themselves. Here’s why.
A note on kernel methods
for readers who want to have a quick reference to kernels + RKHS basics.
A note on learning bounds
a short overview on techniques to prove learning bounds
Learning with biased labels
a review on robust learning with noisy labels
Sparse double descent where network pruning aggravates overfitting
a brief introduction for the ICML paper of sparse double descent