CSC411H1 Study Guide - Final Guide: Gradient Descent, Likelihood Function, Multiclass Classification
Document Summary
Lec 08 mutli-class classification: discriminant functions for k > 2 classes, using binary classifiers. One vs all (ova) classifier use k 1 classifiers. Each classifier solves: is input in class c(, or not. One vs one (ovo) classifier use k k 1 2 = One for each possible pair of classes. Majority vote of all classifier functions classifies input. Problem: some inputs are ambiguous ex. points inside green region: comparison of k-class discriminant vs. ovo, k-class discriminant discriminant use k functions y+, each with an associate vector of weights + + w+,0, y() = argmax y: multi-class linear regression, rewrite k functions into one function of. Input x is assigned to class of the function with the highest output. Decision boundary between class c
