Hardware Acceleration of Deep Learning Algorithms for Real-Time IoT Applications

Authors: Karthik Wali

DOI: https://doi.org/10.37082/IJIRMPS.v7.i6.232595

Short DOI: https://doi.org/

Country: USA

Full-text Research PDF File:   View   |   Download

Abstract: The notion of the IoT has grown exponentially across several areas such as healthcare, the smart world, industrial internetworking, and smart systems. These IoT devices produce tremendous real-time data which needs to be processed to support decision-making on a real-time basis. Such workloads are difficult to solve on traditional computing architectures because they are insufficient in computational capabilities, which brings latencies and high energy consumption. The use of deep learning – a computational process of modeling, analyzing, and understanding unknown and intricate data patterns that enhances the intelligence of a system, is advantageous in IoT applications. However, applying these models to IoT devices is still a major challenge, especially because of the hardware limitations of such devices, memory limitations and power constraints. These are some of the limitations that call for more advanced solutions that can boost the effectiveness and efficiency of deep learning in the IoT context.
Hardware acceleration has emerged as a solution that can be used to fill this gap since it can be implemented through the use of special processing units such as FPGAs, GPUs, and ASICs. Such specialized accelerators can support parallel computation, efficient memory management, and low energy consumption that are needed for DL these models in real-time. This technique helps Internet things to considerably reduce inference time, and energy consumption, and enhance its overall performance by granting those extra resources to computer hardware. Moreover, other computational procedures such as quantization and pruning are other ways of optimizing the model that also improve the possibility of deep learning implementation in edge devices. So as the technology advances even more, what we will see is that hardware accelerators supporting deep learning will act as invaluable enablers for optimizing IoT systems and making them intelligent enough to conduct analysis and draw decisions in real-time.

Keywords:

Paper Id: 232595

Published On: 2019-12-06

Published In: Volume 7, Issue 6, November-December 2019

Share this