How can an explicit approximate kernel map let SGD learn nonlinear boundaries?
Classic MLLevel: IntermediateNonlinear learning with linear solvers
How can an explicit approximate kernel map let SGD learn nonlinear boundaries?
Hints
- Linear in transformed features need not mean linear in raw inputs.
- Approximation replaces implicit kernel computation.
Answer
Transform inputs into nonlinear features whose inner products approximate a kernel, then fit a linear model there. The resulting predictor is nonlinear in original inputs without requiring a full exact kernel matrix.
Source and licence: scikit-learn 1.7.2 — kernel approximation · BSD-3-Clause · answer: Quantoros editorial