Reject Illegal Inputs: Scaling Generative Classifiers with Supervised Deep Infomax

Sep 25, 2019 Blind Submission readers: everyone Show Bibtex
  • Keywords: generative classifiers, selective classification, classification with rejection
  • TL;DR: scale generative classifiers on complex datasets, and evaluate their effectiveness to reject illegal inputs including out-of-distribution samples and adversarial examples.
  • Abstract: Deep Infomax~(DIM) is an unsupervised representation learning framework by maximizing the mutual information between the inputs and the outputs of an encoder, while probabilistic constraints are imposed on the outputs. In this paper, we propose Supervised Deep InfoMax~(SDIM), which introduces supervised probabilistic constraints to the encoder outputs. The supervised probabilistic constraints are equivalent to a generative classifier on high-level data representations, where class conditional log-likelihoods of samples can be evaluated. Unlike other works building generative classifiers with conditional generative models, SDIMs scale on complex datasets, and can achieve comparable performance with discriminative counterparts. With SDIM, we could perform \emph{classification with rejection}. Instead of always reporting a class label, SDIM only makes predictions when test samples' largest logits surpass some pre-chosen thresholds, otherwise they will be deemed as out of the data distributions, and be rejected. Our experiments show that SDIM with rejection policy can effectively reject illegal inputs including out-of-distribution samples and adversarial examples.
  • Original Pdf:  pdf
0 Replies