Outline • Parameter estimation – continued – Non-parametric methods Maximum-Likelihood Estimation • Assumptions – We separate a collection of samples according to class D1, D2, ....., Dc – Samples in Dj are drawn independently according to the probability p(x|wj) – We assume that p(x|wj) has a known parametric form and is uniquely determined by the value of a parameter vector j – To simplify further, we assume that samples in Di give no information about j if i j 5/29/2016 Visual Perception Modeling 2 Maximum-Likelihood Estimation – cont. • Suppose that D contains n samples – x1, ....., xn – By assumption that samples were drawn independently, we have n p ( D | θ ) p ( xk | θ ) k 1 – The maximum-likelihood estimate of is the value of * that maximizes p(D| ) 5/29/2016 Visual Perception Modeling 3 Bayesian Estimation • Assumptions – The form of the density p(x|) is assumed to be known, but the value of the parameter vector is not known exactly – Our initial knowledge about is assumed to be contained in a known prior density p() – The rest of our knowledge about is contained in a set D of n samples x1, ....., xn drawn independently according to the unknown probability density p(x) 5/29/2016 Visual Perception Modeling 4 Bayesian Estimation – cont. • General theory – The basic problem is to compute the posterior density p(|D) – By Bayes formula we have p( | D) p( D | ) p( ) p( D | ) p( )d – By the independence assumption n p ( D | ) p ( xk | ) k 1 5/29/2016 Visual Perception Modeling 5 Bayesian Estimation – cont. • Gaussian case – The univariate case p(m|D) 5/29/2016 Visual Perception Modeling 6 Bayesian Estimation – cont. • Gaussian case – continued – The univariate case p(x|D) – The multivariate case 5/29/2016 Visual Perception Modeling 7 Non-parametric Methods • In maximum-likelihood and Bayesian estimation – The forms of the probability densities are assumed to be known – However, the assumed forms rarely fit the densities in practice • In particular, all of the classical parametric densities are uni-modal 5/29/2016 Visual Perception Modeling 8 A Multimodal Density 5/29/2016 Visual Perception Modeling 9 Solutions • More complicated parametric models – Mixture of Gaussians – More general, some basis functions to describe a probability density – Learning is intrinsically more difficult when we have more parameters • Non-parametric methods 5/29/2016 Visual Perception Modeling 10 Non-parametric Methods • Most of the non-parametric density estimation methods are based on the following fact – The probability P that a vector x will fall in a region R is given by P p( x)dx R 5/29/2016 Visual Perception Modeling 11 Non-parametric Methods – cont. • For n smaples x1, ....., xn that are drawn independently according to p(x), the probability that k of n will be in R is given by n k n k Pk P (1 P) [k] nP k P p( x)dx p( x)V R k/n p( x) V 5/29/2016 Visual Perception Modeling V is the volume of R 12 Non-parametric Methods – cont. 5/29/2016 Visual Perception Modeling 13 Non-parametric Methods – cont. • Problems to be addressed – If we fix the volume V and have more samples, the ratio k/n will converge as desired • Averaged version of p(x) – How to estimate p(x)? • Let V approach zero? 5/29/2016 Visual Perception Modeling 14 Parzen Windows • Parzen windows – We use a window function for interpolation, each sample contributing to the estimate in accordance with its distance from x 1 n 1 x xi pn ( x) n i 1 Vn hn • Here hn is a parameter 5/29/2016 Visual Perception Modeling 15 Parzen Windows - cont. • Choice of hn – Too large, the spatial resolution is low – Too small, the estimate will have a large variance 5/29/2016 Visual Perception Modeling 16 Parzen Windows - cont. • Properties – Convergence of mean • As n approaches infinity, the estimate will also approach p(x) if p(x) is continuous • Smaller Vn is better – Convergence of variance • A smaller variance needs a larger Vn 5/29/2016 Visual Perception Modeling 17 Parzen Windows - cont. 5/29/2016 Visual Perception Modeling 18 Parzen Windows - cont. 5/29/2016 Visual Perception Modeling 19 Parzen Windows - cont. 5/29/2016 Visual Perception Modeling 20 Parzen Windows - cont. 5/29/2016 Visual Perception Modeling 21 Kn-Nearest-Neighbor Estimation • Let the cell volume be a function of the training data – To estimate p(x) from n samples, we can center a cell about x and let it grow until it captures kn samples kn / n pn ( x ) Vn 5/29/2016 Visual Perception Modeling 22 Kn-Nearest-Neighbor Estimation – cont. 5/29/2016 Visual Perception Modeling 23 Kn-Nearest-Neighbor Estimation – cont. 5/29/2016 Visual Perception Modeling 24 Kn-Nearest-Neighbor Estimation – cont. 5/29/2016 Visual Perception Modeling 25 The Nearest-Neighbor Rule • The nearest-neighbor rule – Let Dn={x1, ...., xn} denote a set of n labeled prototypes – Suppose that x' be the prototype nearest to a test point x – We classify x to the class associated with x' 5/29/2016 Visual Perception Modeling 26 The Nearest-Neighbor Rule – cont. 5/29/2016 Visual Perception Modeling 27