A basis function is just a function from R^n -> R, and if you combine a bunch of them (each one applied to the same input vector) you get a new space where classification may be easier (see e.g. fig 1 in this wonderful paper: http://cbcl.mit.edu/publications/ps/serre-poggio_%20ANeuromorphicApproachtoComputerVision_CACM_Oct2010.pdf ).
Can be learnt from data by e.g. neural network (each hidden unit learns a basis function), Boltzmann machines, and others.
A basis function is just a function from R^n -> R, and if you combine a bunch of them (each one applied to the same input vector) you get a new space where classification may be easier (see e.g. fig 1 in this wonderful paper: http://cbcl.mit.edu/publications/ps/serre-poggio_%20ANeuromorphicApproachtoComputerVision_CACM_Oct2010.pdf ).
ReplyDeleteCan be learnt from data by e.g. neural network (each hidden unit learns a basis function), Boltzmann machines, and others.