http:/parnec.nuaa.edu.cn Towards a theory of meta-features Theorem 1 Let h be the class of linear discrimina tion functions on Rd. We assume that the hyperplane xn ngy=hw(x) passes thr rough the origin, i.e. h E H, hw(x) =9(u) ear regression functions from meta-features to weights RN→R,ie,vg∈g,gu;c)=cu, uphere c is a g) vector of k parameters. Then VCD(Hg)smin(k, d)Company name www.themegallery.com Towards a theory of meta-features