Loss for multiclass classification

• Map $x$ to one of $K>2$ classes
• Distribution - Categorical Distribution
• The parameters are constrained to take values between zero and one, and they must collectively sum to one to ensure a valid probability distribution $Pr(y=k)={\lambda }_k$
• Network computes $K$ params from input x
• Since the output of the network does not conform to the format, we use Softmax so results are positive, and the K numbers sum to one
• 
• Loss function : Negative Log Likelihood
• This is then - Cross Entropy
•