Regression NSS: An Alternative to Cross Validation

The Noise Sensitivity Signature (NSS), originally introduced by Grossman and Lapedes (1993), was proposed as an alternative to cross validation for selecting network complexity. In this paper, we extend NSS to the general problem of regression estimation. We also present results from regularized linear regression simulations which indicate that for problems with few data points, NSS regression estimates perform better than Generalized Cross Validation (GCV) regression estimates [7]. 1 The Noise Sensitivity Signature (NSS) One of the fundamental problems in neural networks and regression estimation in general is to develop reliable methods for avoiding overfitting to finite data. Grossman and Lapedes (1993) proposed the NSS for classifications problems as a method for selecting neural network complexity to match the complexity of the data available and in this way reduce overfitting. Unlike other methods for avoiding overfitting, such as cross validation [5], which uses only part ...

• Type: Conference
• Year: 1995
• Where: COLT
• Authors: Michael P. Perrone, Brian S. Blais
• Keywords: COLT 1995, Cross Validation, Machine Learning, Network Complexity, Noise Sensitivity Signature