Optimizing convolutional neural networks for fast inference has recently become an extremely active area of research. One of the go-to solutions in this context is weight pruning, which aims to reduce computational and memory footprint by removing large subsets of the connections in a neural network. Surprisingly, much less attention has been given to exploiting sparsity in the activation maps, which tend to be naturally sparse in many settings thanks to the structure of rectified linear (ReLU) activation functions. In this paper, we present an in-depth analysis of methods for maximizing the sparsity of the activations in a trained neural network, and show that, when coupled with an efficient sparse-input convolution algorithm, we can leverage this sparsity for significant performance gains. To induce highly sparse activation maps without accuracy loss, we introduce a new regularization technique, coupled with a new threshold-based sparsification method based on a parameterized activation function called Forced-Activation-Threshold Rectified Linear Unit (FATReLU). We examine the impact of our methods on popular image classification models, showing that most architectures can adapt to significantly sparser activation maps without any accuracy loss. Our second contribution is showing that these these compression gains can be translated into inference speedups: we provide a new algorithm to enable fast convolution operations over networks with sparse activations, and show that it can enable significant speedups for end-to-end inference on a range of popular models on the large-scale ImageNet image classification task on modern Intel CPUs, with little or no retraining cost.

Convergence of critical infrastructure and data, including government and enterprise, to the dynamic Internet of Things (IoT) environment and future digital ecosystems exhibit significant challenges for privacy and identity in these interconnected domains. There are an increasing variety of devices and technologies being introduced, rendering existing security tools inadequate to deal with the dynamic scale and varying actors. The IoT is increasingly data driven with user sovereignty being essential - and actors in varying scenarios including user/customer, device, manufacturer, third party processor, etc. Therefore, flexible frameworks and diverse security requirements for such sensitive environments are needed to secure identities and authenticate IoT devices and their data, protecting privacy and integrity. In this paper we present a review of the principles, techniques and algorithms that can be adapted from other distributed computing paradigms. Said review will be used in application to the development of a collaborative decision-making framework for heterogeneous entities in a distributed domain, whilst simultaneously highlighting privacy preserving issues in the IoT. In addition, we present our trust-based privacy preserving schema using Dempster-Shafer theory of evidence. While still in its infancy, this application could help maintain a level of privacy and nonrepudiation in collaborative environments such as the IoT.