Andrew Howard, Tony Jebara

Poster M29

Learning Monotonic Transformations for Classification
We simultaneously learn a maximum margin classifier and an arbitrary monotonic transformation T ( x ) of the inputs in a convex framework using a semidefinite relaxation. D  y = sign   wd T ( xd ) + b   d =1 
x1 x2 xD
WebKB error

w1 w2 wD b
y

Example: Gender Classification. Learn intensity transfer function for optimal linear classification.

Corel error

T ( x)

Gender error