Statistical Learning Theory and Algorithms
John O. Moody · 1993
This project addressed three fundamental areas in statistical learning theory and algorithms: (1) a practical and theoretically sound method or estimating generalization performance of nonlinear learning systems (Generalized Prediction Error, GPE), (2) a more powerful and efficient class of network architectures (Parameterized Projection Pursuit Regression (P(3)R) networks), and (3) faster real-time learning methods based on asymptotically optimal stochastic gradient search. Three papers were published under this grant. Additionally, a graduate student finished his PhD under research topic Networks with Learned Unit Response Functions .