Outline • Classification Bayesian Decision Rule • A two-class example – 1 for sea bass – 2 for salmon • Prior probability – P(1) – P(2) 5/29/2016 Visual Perception Modeling 2 Bayesian Decision Rule – cont. • Class conditional probability density – P(1 | x) – P(2 | x) • Bayes formula p( x | i ) P( i ) P( i ) P( x) 5/29/2016 Visual Perception Modeling 3 Bayesian Decision Rule – cont. • Bayes decision rule – Decide 1 if P(1 | x) > P(2 | x) – Otherwise decide 2 – The optimal decision rule • Minimize the average error we make 5/29/2016 Visual Perception Modeling 4 Feature Space • Feature space – The Euclidean space Rd if we use a ddimensional feature – Each possible feature is a point the Rd space 5/29/2016 Visual Perception Modeling 5 Loss Function • Loss function – States exactly how costly each action is – Is used to convert a probability determination into a decision – Allows us to treat situations where some kinds of classification mistakes are more costly than others • Equally costly is a special case 5/29/2016 Visual Perception Modeling 6 Loss Function – cont. • Suppose that there are c categories – {1, 2, ....., c} • There are a possible actions – {1, 2, ....., c} • Loss function (i | j} describe the loss incurred for taking action i when the state of nature is j 5/29/2016 Visual Perception Modeling 7 Loss Function – cont. • Bayes formula p( x | i ) P( i ) P( i | x) P( x) – where c p( x) p( x | j ) P( j ) j 1 5/29/2016 Visual Perception Modeling 8 Loss Function – cont. • The expected loss function given a particular observation x c R( i | x) ( i | j ) P( j | x) • The overall risk j 1 R R( ( x) | x) p( x)dx 5/29/2016 Visual Perception Modeling 9 Bayes Decision Rule • To minimize the overall risk, compute the conditional risk and select the action for the conditional risk is minimum c R( i | x) ( i | j ) P( j | x) j 1 – The resulting minimum overall risk is called the Bayes risk, which is the best performance 5/29/2016 Visual Perception Modeling 10 Two-Category Classification • Two categories • Two actions • We decide 1 if (21 11) p( x | 1 ) P(1 ) (12 22 ) p( x | 2 ) P(2 ) 5/29/2016 Visual Perception Modeling 11 Minimum-Error-Rate Classification • Zero-one loss • For minimum error rate, – Decide 1 if P(1 | x) > P(2 | x) – This is the Bayes decision rule 5/29/2016 Visual Perception Modeling 12 Discriminant Functions • The classifier is said to assign a feature vector x to class i if – gi(x) > gj(x) for all j i – This can be viewed as a network – If f(.) is a monotonically increasing function, f(g(x)) and g(x) as discriminant function will give the same classification result 5/29/2016 Visual Perception Modeling 13 Decision Regions • The effect of decision rule is to divide the feature space into c decision regions – R1, R2, ...., Rc – The regions are separated by decision boundaries – Two-category case 5/29/2016 Visual Perception Modeling 14 Normal Density • Gaussian density p ( x) 1 (2 ) d / 2 | |1/ 2 1 t 1 exp ( x ) ( x ) 2 – Properties • • • • 5/29/2016 Mean Variance Entropy Central limit theorem Visual Perception Modeling 15 Discriminant Functions for Normal Density • Minimum error rate classification for normal density 1 d t 1 g i ( x) ( x i ) i ( x i ) ln( 2 ) 2 2 1 ln(| i |) ln( P( i )) 2 – Special cases 5/29/2016 Visual Perception Modeling 16