Introduction to Nonlinear Statistics and Neural Networks：介绍非线性统计和神经网络.ppt

3/11/2009 Meto 630; V.Krasnopolsky, Nonlinear Statistics and NNs Introduction to Nonlinear Statistics and Neural Networks Vladimir Krasnopolsky ESSIC/UMD NCEP/NOAA (SAIC) /mmab/people/kvladimir.html Outline Introduction: Regression Analysis Regression Models (Linear Nonlinear) NN Tutorial Some Atmospheric Oceanic Applications Accelerating Calculations of Model Physics Developing New NN Parameterizations How to Apply NNs Conclusions Evolution of Statistical Methods Problems for Classical Paradigm: Nonlinearity Complexity High Dimensionality - Curse of Dimensionality New Paradigm under Construction: Is still quite fragmentary Has many different names and gurus NNs are one of the tools developed inside this paradigm Statistical Inference: A Generic Problem Problem: Information exists in the form of finite sets of values of several related variables (sample or training set) – a part of the population: = {(x1, x2, ..., xn)p, zp}p=1,2,...,N x1, x2, ..., xn - independent variables (accurate), z - response variable (may contain observation errors ε) We want to find responses z’q for another set of independent variables = {(x’1, x’2, ..., x’n)q}q=1,..,M Regression Analysis (1):General Solution and Its Limitations Regression Analysis (2):A Generic Solution The effect of independent variables on the response is expressed mathematically be the regression or response function f: y = f( x1, x2, ..., xn; a1, a2, ..., aq) y - dependent variable a1, a2, ..., aq - regression parameters (unknown!) f - the form is usually assumed to be known Regression model for observed response variable: Z = y + ε = f(x1, x2, ..., xn; a1, a2, ..., aq) + ε ε - error in observed value z Regression Models (1):Maximum Likelihood Fischer suggested to determine unknown regression parameters {ai}i=1,..,q maximizing the functional: here ρ(ε) is the probability density function of errors εi In a case when ρ(ε