Diffy: a Déjà vu-Free Differential Deep Neural Network Accelerator
Authors:
- Mostafa MahmoudMMMostafa Mahmoud
- Kevin SiuKSKevin Siu
- Andreas MoshovosAMAndreas Moshovos
- 1 October 2018
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 134-147
Abstract
We show that Deep Convolutional Neural Network (CNN) implementations of computational imaging tasks exhibit spatially correlated values. We exploit this correlation to reduce the amount of computation, communication, and storage needed to execute such CNNs by introducing Diffy, a hardware accelerator that performs Differential Convolution. Diffy stores, communicates, and processes the bulk of the activation values as deltas. Experiments show that, over five state-of-the-art CNN models and for HD resolution inputs, Diffy boosts the average performance by 7.1× over a baseline value-agnostic accelerator [1] and by 1.41× over a state-of-the-art accelerator that processes only the effectual content of the raw activation values [2]. Further, Diffy is respectively 1.83× and 1.36× more energy efficient when considering only the on-chip energy. However, Diffy requires 55% less on-chip storage and 2.5× less off-chip bandwidth compared to storing the raw values using profiled per-layer precisions [3]. Compared to using dynamic per group precisions [4], Diffy requires 32% less storage and 1.43× less off-chip memory bandwidth. More importantly, Diffy provides the performance necessary to achieve real-time processing of HD resolution images with practical configurations. Finally, Diffy is robust and can serve as a general CNN accelerator as it improves performance even for image classification models.Keywords
This publication has 38 references indexed in Scilit:
- Learning to DeblurIEEE Transactions on Pattern Analysis and Machine Intelligence, 2015
- Fully convolutional networks for semantic segmentation2015
- Image Super-Resolution Using Deep Convolutional NetworksIEEE Transactions on Pattern Analysis and Machine Intelligence, 2015
- The Noise Clinic: a Blind Image Denoising AlgorithmImage Processing On Line, 2015
- DaDianNao: A Machine-Learning Supercomputer2014
- Weighted Nuclear Norm Minimization with Application to Image Denoising2014
- Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation2014
- On Single Image Scale-Up Using Sparse-Representations2012
- Fields of Experts: A Framework for Learning Image Priors2005
- A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics2002