Your Classifier is Secretly an Energy Based Model and You Should Treat\n it Like One

We propose to reinterpret a standard discriminative classifier of p(y|x) as\nan energy based model for the joint distribution p(x,y). In this setting, the\nstandard class probabilities can be easily computed as well as unnormalized\nvalues of p(x) and p(x|y). Within this framework, standard discriminative\narchitectures may beused and the model can also be trained on unlabeled data.\nWe demonstrate that energy based training of the joint distribution improves\ncalibration, robustness, andout-of-distribution detection while also enabling\nour models to generate samplesrivaling the quality of recent GAN approaches. We\nimprove upon recently proposed techniques for scaling up the training of energy\nbased models and presentan approach which adds little overhead compared to\nstandard classification training. Our approach is the first to achieve\nperformance rivaling the state-of-the-artin both generative and discriminative\nlearning within one hybrid model.\n