A Neyman-Pearson approach to statistical learning Export

@article{citeulike:600676, abstract = {The Neyman-Pearson (NP) approach to hypothesis testing is useful in situations where different types of error have different consequences or a priori probabilities are unknown. For any /spl alpha/>0, the NP lemma specifies the most powerful test of size /spl alpha/, but assumes the distributions for each hypothesis are known or (in some cases) the likelihood ratio is monotonic in an unknown parameter. This paper investigates an extension of NP theory to situations in which one has no knowledge of the underlying distributions except for a collection of independent and identically distributed (i.i.d.) training examples from each hypothesis. Building on a "fundamental lemma" of Cannon et al., we demonstrate that several concepts from statistical learning theory have counterparts in the NP context. Specifically, we consider constrained versions of empirical risk minimization (NP-ERM) and structural risk minimization (NP-SRM), and prove performance guarantees for both. General conditions are given under which NP-SRM leads to strong universal consistency. We also apply NP-SRM to (dyadic) decision trees to derive rates of convergence. Finally, we present explicit algorithms to implement NP-SRM for histograms and dyadic decision trees.}, author = {Scott, C. and Nowak, R.}, citeulike-article-id = {600676}, citeulike-linkout-0 = {http://dx.doi.org/10.1109/TIT.2005.856955}, citeulike-linkout-1 = {http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1522642}, doi = {10.1109/TIT.2005.856955}, journal = {Information Theory, IEEE Transactions on}, keywords = {erm, information, neyman-pearson, srm, statistics, theory}, number = {11}, pages = {3806--3819}, posted-at = {2008-06-26 13:10:36}, priority = {2}, title = {A Neyman-Pearson approach to statistical learning}, url = {http://dx.doi.org/10.1109/TIT.2005.856955}, volume = {51}, year = {2005} }

TY - JOUR ID - citeulike:600676 L3 - citeulike-article-id:600676 N2 - The Neyman-Pearson (NP) approach to hypothesis testing is useful in situations where different types of error have different consequences or a priori probabilities are unknown. For any /spl alpha/>0, the NP lemma specifies the most powerful test of size /spl alpha/, but assumes the distributions for each hypothesis are known or (in some cases) the likelihood ratio is monotonic in an unknown parameter. This paper investigates an extension of NP theory to situations in which one has no knowledge of the underlying distributions except for a collection of independent and identically distributed (i.i.d.) training examples from each hypothesis. Building on a "fundamental lemma" of Cannon et al., we demonstrate that several concepts from statistical learning theory have counterparts in the NP context. Specifically, we consider constrained versions of empirical risk minimization (NP-ERM) and structural risk minimization (NP-SRM), and prove performance guarantees for both. General conditions are given under which NP-SRM leads to strong universal consistency. We also apply NP-SRM to (dyadic) decision trees to derive rates of convergence. Finally, we present explicit algorithms to implement NP-SRM for histograms and dyadic decision trees. IS - 11 JF - Information Theory, IEEE Transactions on EP - 3819 TI - A Neyman-Pearson approach to statistical learning VL - 51 SP - 3806 KW - erm KW - information KW - neyman-pearson KW - srm KW - statistics KW - theory AU - Scott, C AU - Nowak, R PY - 2005/// UR - http://dx.doi.org/10.1109/TIT.2005.856955 ER -