Prior Activation Distribution (PAD): A Versatile Representation to Utilize DNN Hidden Units
This provides a method for improving uncertainty estimation and out-of-distribution detection in neural networks, but it is incremental as it builds on existing activation analysis techniques.
The paper tackles the problem of capturing activation patterns in deep neural networks for classification by introducing Prior Activation Distribution (PAD), a technique that uses statistical measures to derive uncertainty estimates, achieve competitive inference accuracy, and isolate out-of-distribution samples on benchmark datasets like MNIST and CIFAR10.
In this paper, we introduce the concept of Prior Activation Distribution (PAD) as a versatile and general technique to capture the typical activation patterns of hidden layer units of a Deep Neural Network used for classification tasks. We show that the combined neural activations of such a hidden layer have class-specific distributional properties, and then define multiple statistical measures to compute how far a test sample's activations deviate from such distributions. Using a variety of benchmark datasets (including MNIST, CIFAR10, Fashion-MNIST & notMNIST), we show how such PAD-based measures can be used, independent of any training technique, to (a) derive fine-grained uncertainty estimates for inferences; (b) provide inferencing accuracy competitive with alternatives that require execution of the full pipeline, and (c) reliably isolate out-of-distribution test samples.