Sparsity in penalized empirical risk minimization
Koltchinskii, Vladimir
Ann. Inst. H. Poincaré Probab. Statist., Tome 45 (2009) no. 1, p. 7-57 / Harvested from Project Euclid
Let (X, Y) be a random couple in S×T with unknown distribution P. Let (X1, Y1), …, (Xn, Yn) be i.i.d. copies of (X, Y), Pn being their empirical distribution. Let h1, …, hN:S↦[−1, 1] be a dictionary consisting of N functions. For λ∈ℝN, denote fλ:=∑j=1Nλjhj. Let ℓ:T×ℝ↦ℝ be a given loss function, which is convex with respect to the second variable. Denote (ℓ•f)(x, y):=ℓ(y; f(x)). We study the following penalized empirical risk minimization problem ¶ \[\hat{\lambda}^{\varepsilon }:=\mathop{\operatorname {argmin}}_{\lambda\in {\mathbb{R}}^{N}}\bigl[P_{n}(\ell\bullet f_{\lambda})+\varepsilon \|\lambda\|_{\ell_{p}}^{p}\bigr],\] ¶ which is an empirical version of the problem ¶ \[\lambda^{\varepsilon }:=\mathop{\operatorname{argmin}}_{\lambda\in {\mathbb{R}}^{N}}\bigl[P(\ell \bullet f_{\lambda})+\varepsilon \|\lambda\|_{\ell_{p}}^{p}\bigr]\] ¶ (here ɛ≥0 is a regularization parameter; λ0 corresponds to ɛ=0). A number of regression and classification problems fit this general framework. We are interested in the case when p≥1, but it is close enough to 1 (so that p−1 is of the order $\frac{1}{\log N}$ , or smaller). We show that the “sparsity” of λɛ implies the “sparsity” of λ̂ɛ and study the impact of “sparsity” on bounding the excess risk P(ℓ•fλ̂ɛ)−P(ℓ•fλ0) of solutions of empirical risk minimization problems.
Publié le : 2009-02-15
Classification:  Empirical risk,  Penalized empirical risk,  ℓ_p-penalty,  Sparsity,  Oracle inequalities,  62G99,  62J99,  62H30
@article{1234469970,
     author = {Koltchinskii, Vladimir},
     title = {Sparsity in penalized empirical risk minimization},
     journal = {Ann. Inst. H. Poincar\'e Probab. Statist.},
     volume = {45},
     number = {1},
     year = {2009},
     pages = { 7-57},
     language = {en},
     url = {http://dml.mathdoc.fr/item/1234469970}
}
Koltchinskii, Vladimir. Sparsity in penalized empirical risk minimization. Ann. Inst. H. Poincaré Probab. Statist., Tome 45 (2009) no. 1, pp.  7-57. http://gdmltest.u-ga.fr/item/1234469970/