DeepRecon: Dynamically reconfigurable architecture for accelerating deep neural networks

Deep learning models are computationally expensive and their performance depends strongly on the underlying hardware platform. General purpose compute platforms such as GPUs have been widely used for implementing deep learning techniques. However, with the advent of emerging application domains such as internet of things, developments of custom integrated circuits capable of efficiently implementing deep learning models with low power and form factor are in high demand. In this paper we analyze both the computation and communication costs of common deep networks. We propose a reconfigurable architecture that efficiently utilizes computational and storage resources for accelerating deep learning techniques without loss of algorithmic accuracy.

1. Introduction

Mobile devices are becoming increasingly involved in our daily lives as diverse and active participants in our decision making. Using speech recognition and understanding at the level of intelligent response and language translation, visual recognition and scene understanding to take signals from the environment, and smart search of the vast knowledge database on the cloud, our devices are becoming sophisticated partners in our interactions. The majority of data analysis and processing that goes into supporting these systems resides in warehouse-scale computers and is communicated via the cloud infrastructure. Moving some of the processing from the cloud to the device would make our devices more intelligent, responsive and would save a lot of energy. That would require improving the efficiency of both the algorithms and the underlying mobile platform that performs the processing.