We propose a method to estimate the uncertainty of the outcome of an image classifier on a given input datum. Deep neural networks commonly used for image classification are deterministic maps from an input image to an output class. As such, their outcome on a given datum involves no uncertainty, so we must specify what variability we are referring to when defining, measuring and interpreting confidence. To this end, we introduce the Wellington Posterior, which is the distribution of outcomes that would have been obtained in response to data that could have been generated by the same scene that produced the given image. Since there are infinitely many scenes that could have generated the given image, the Wellington Posterior requires induction from scenes other than the one portrayed. We explore alternate methods using data augmentation, ensembling, and model linearization. Additional alternatives include generative adversarial networks, conditional prior networks, and supervised single-view reconstruction. We test these alternatives against the empirical posterior obtained by inferring the class of temporally adjacent frames in a video. These developments are only a small step towards assessing the reliability of deep network classifiers in a manner that is compatible with safety-critical applications.
تحميل البحث