By: A Dahiya, B. Gupta
Artificial intelligence-based services and applications have taken the world by storm as there is a major breakthrough in technologies like IoT and mobile computing contributing to generating zillions of data every day. Though this data has constantly been utilized in improving the performance of deep neural networks; but, the storage and processing of this much data is a serious problem. Further, improved performance of deep neural networks costs increased computational complexity and large resource consumption. Therefore, large deep learning models cannot be deployed at the end devices or mobile devices. Meanwhile, cloud computing has its own constraints like high latency, high bandwidth costs, reliability, etc. Cloud computing is known for scalable resources at reasonable cost, data storage, and network management functions. All these functions are carried out using centralized data centers. However, these centralized data centers fail to provide real-time services to billions of users that are distributed across the globe. In this scenario, edge computing comes to the rescue by providing high computing nodes at the edge of the network and that too with low latency. These features of edge computing can support resource-intensive AI applications on end devices. Since edge computing is nearer to users, it can solve many of the issues related to cloud computing. Figure 1 shows the deployment of deep learning on edge computing.
One of the most widely used AI techniques is deep learning. Deep learning is a multi-level layer structure that imitates the human brain to process the data and generates patterns that would help in decision-making. It is a subset of machine learning in artificial intelligence, having the capability to learn unsupervised from unstructured and disorganized data. With the advancement of the digital age, data explodes in all shapes and sizes and from every corner of the globe, which is referred to as big data. This generated data is so vast and unstructured that humans cannot process it and extract relevant information from it. Self-learning, self-training, adaptive and dynamic features of deep neural networks make it suitable for processing and analyzing big data generated from sources like social media, e-commerce, the internet, search engines, online cinemas, and many more.
Edge intelligence or edge AI is the new emerging era that tends to boost up the AI applications using edge resources. This is a new concept, and it is yet to be explored. There are many advantages associated with pushing AI to the edge nodes. The first is latency which is a major hindrance in cloud computing. Since deep learning models are closer to the requested users, therefore, latency can be significantly reduced. Second is privacy protection, as unstructured data required for processing is stored on edge nodes instead of cloud data centers. Next, edge computing has a hierarchical and decentralized architecture that can provide more reliability to the users. Next, multi-faceted deep learning services can advertise or commercialize growth edge computing.
Deep learning models have proven to be very beneficial in integrating cloud and edge computing seamlessly. Ubiquitous AI is the next milestone to be achieved by major IT giants like Google, Microsoft, IBM, etc. These organizations have the vision to make AI to be available everywhere and to everyone. There is no denying the fact that edge computing can play a key role in achieving this role. The users can rapidly access Real-time data generated by edge devices to train the AI model and infer useful patterns that can help in decision-making and enhance productivity. This concept can face many challenges like performance and energy efficiency issues.
- Chen, J., & Ran, X. (2019). Deep Learning With Edge Computing: A Review. Proc. IEEE, 107(8), 1655-1674.
- Wang, X., Han, Y., Leung, V. C., Niyato, D., Yan, X., & Chen, X. (2020). Convergence of edge computing and deep learning: A comprehensive survey. IEEE Communications Surveys & Tutorials, 22(2), 869-904.
Cite this article:
A Dahiya, B. Gupta (2021), Edge Intelligence: A New Emerging Era, Insights2Techinfo, pp.1