This function computes the derivative of the function $$v\mapsto \frac{w}{\|w\|_A}$$ with respect to y.
dA(w, A, dw)
w | vector of length n. |
---|---|
A | square matrix that defines the norm |
dw | derivative of w with respect to y. As y is a vector of length n, the derivative is a matrix of size nxn. |
the Jacobian matrix of the normalization function. This is a matrix of size nxn.
The first derivative of the normalization operator is $$\frac{\partial}{\partial y}\left(w\mapsto \frac{w}{\|w\|_A}\right)=\frac{1}{\|w\|}\left(I_n - \frac{w w^ \top A}{w^\top w}\right) \frac{\partial w}{\partial y}$$
Kraemer, N., Sugiyama M. (2011). "The Degrees of Freedom of Partial Least Squares Regression". Journal of the American Statistical Association 106 (494) https://www.tandfonline.com/doi/abs/10.1198/jasa.2011.tm10107
Kraemer, N., Braun, M.L. (2007) "Kernelizing PLS, Degrees of Freedom, and Efficient Model Selection", Proceedings of the 24th International Conference on Machine Learning, Omni Press, 441 - 448
Nicole Kraemer