Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
This work addresses efficiency for CNN deployment, particularly in resource-constrained environments, but is incremental as it builds on existing spatial correlation observations.
The paper tackles the high computational demands of convolutional neural networks (CNNs) by proposing a method to predict zero-valued activations in output feature maps, reducing convolution operations and achieving up to 30% MAC reduction with minimal accuracy loss.
Convolutional neural networks (CNNs) introduce state-of-the-art results for various tasks with the price of high computational demands. Inspired by the observation that spatial correlation exists in CNN output feature maps (ofms), we propose a method to dynamically predict whether ofm activations are zero-valued or not according to their neighboring activation values, thereby avoiding zero-valued activations and reducing the number of convolution operations. We implement the zero activation predictor (ZAP) with a lightweight CNN, which imposes negligible overheads and is easy to deploy on existing models. ZAPs are trained by mimicking hidden layer ouputs; thereby, enabling a parallel and label-free training. Furthermore, without retraining, each ZAP can be tuned to a different operating point trading accuracy for MAC reduction.