Outline • Parameter estimation – continued – Bayesian learning Announcement • This coming Friday, March 2, there will be no class during normal class time 10:1011AM. • The class will be moved to 12:20-1:10PM in Room 499, Dirac Science library • Homework #2 is due today 5/29/2016 Visual Perception Modeling 2 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 3 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 4 Maximum-Likelihood Estimation – cont. • Log-likelihood l (θ ) ln( p ( D | θ )) θ* arg max l (θ) n θ l (θ) ln( p ( xk | θ)) k 1 n θ l (θ) θ (ln( p ( xk | θ)) ) k 1 5/29/2016 Visual Perception Modeling 5 Maximum-Likelihood Estimation – cont. • The maximum likelihood solution is θl (θ) 0 – A solution * can be a true global maximum, a local maximum, or a minimum, or an inflection point of l() • We need to check each solution individually • Or calculate the second derivatives to identify the global optimum 5/29/2016 Visual Perception Modeling 6 Bayesian Estimation • We assume that parameters are random variables – Training data allows us to convert a distribution on the parameters into a posterior probability density – This is conceptually very different from maximum-likelihood estimation • There the parameters are fixed but not known 5/29/2016 Visual Perception Modeling 7 Bayesian Estimation – cont. • 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 8 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 9 Bayesian Estimation – cont. 5/29/2016 Visual Perception Modeling 10 Bayesian Estimation – cont. • Recursive Bayesian learning – Let Dn={x1, ....., xn} – We have for n > 1 p( | D n ) p( xn | ) p( | D n 1 ) n 1 p ( x | ) p ( | D )d n – Incremental or on-line learning • Learning goes on as the data are collected 5/29/2016 Visual Perception Modeling 11 Bayesian Estimation – cont. 5/29/2016 Visual Perception Modeling 12 Bayesian Estimation – cont. • Bayesian learning – As n increases, p(|Dn) becomes shaper and shaper, approaching a Dirac delta function as n goes to infinity. This behavior is commonly known as Bayesian learning • Relation to the maximum-likelihood solution – If p(|Dn) is very peaky, then the solution from the maximum-likelihood estimate would be close to the true solution 5/29/2016 Visual Perception Modeling 13 Bayesian Estimation – cont. • Why Bayesian estimation – A measure of uncertainty about the current estimation 5/29/2016 Visual Perception Modeling 14 Bayesian Estimation – cont. • Why Bayesian estimation - continued – Learning recursively or incrementally – Use all the available information to compute the desired density p(x|D) p( x | D) p( x | ) p( | D)d 5/29/2016 Visual Perception Modeling 15 Bayesian Estimation – cont. • Gaussian case – The univariate case p(m|D) 5/29/2016 Visual Perception Modeling 16 Bayesian Estimation – cont. • Gaussian case – continued – The univariate case p(x|D) – The multivariate case 5/29/2016 Visual Perception Modeling 17