Unsupervised Energy-based Out-of-distribution Detection using Stiefel-Restricted Kernel Machine

Detecting out-of-distribution (OOD) samples is an essential requirement for\nthe deployment of machine learning systems in the real world. Until now,\nresearch on energy-based OOD detectors has focused on the softmax confidence\nscore from a pre-trained neural network classifier with access to class labels.\nIn contrast, we propose an unsupervised energy-based OOD detector leveraging\nthe Stiefel-Restricted Kernel Machine (St-RKM). Training requires minimizing an\nobjective function with an autoencoder loss term and the RKM energy where the\ninterconnection matrix lies on the Stiefel manifold. Further, we outline\nmultiple energy function definitions based on the RKM framework and discuss\ntheir utility. In the experiments on standard datasets, the proposed method\nimproves over the existing energy-based OOD detectors and deep generative\nmodels. Through several ablation studies, we further illustrate the merit of\neach proposed energy function on the OOD detection performance.\n