Conditional KRR: Injecting Unpenalized Features into Kernel Methods with Applications to Kernel Thresholding
Researchers have developed a method called conditional kernel ridge regression (conditional KRR) that enhances kernel methods by incorporating unpenalized features. This approach is analogous to performing standard linear regression on a function class $\mathcal{F}$ and then applying KRR to the remaining unexplained variance. Theoretical analysis shows that conditional KRR can outperform standard KRR when the $\mathcal{F}$-component of the regression function is significant, a finding supported by experimental results. AI
IMPACT Introduces a novel kernel method that may offer improved performance in specific regression tasks.