Exercise 12.6 - Derivation of Fisher's linear discriminant

Answers

This is but straightforward algebra:

∂𝐽 𝐰 = 𝐒 B 𝐰 𝐰 T 𝐒 W 𝐰 𝐰 T 𝐒 B 𝐰 𝐒 W 𝐰 ( 𝐰 T 𝐒 W 𝐰 ) 2 = 1 ( 𝐰 T 𝐒 W 𝐰 ) 2 𝐰 T 𝐒 W 𝐰 ( 𝐒 B 𝐰 𝐰 T 𝐒 B 𝐰 𝐰 T 𝐒 W 𝐰 𝐒 W 𝐰 ) .

Note that minimizing the Fisher loss is but one option in maximizing 𝐰 T 𝐒 B 𝐰 while minimizing 𝐰 T 𝐒 W 𝐰 . Modern numerical toolkits offer more choices to be explored.

User profile picture
2021-03-24 13:42
Comments