a loss function tells how good our current classifier is You tell your algorithm what kind of errors you care about and what kind of errors you trade off against Multi-class SVM loss –– j could be the number of classes our dataset have -syi – the score of the true class/ s- predicted scores come out