Many of the gateways currently used for connecting IoT devices have some processing capabilities. Limited machine learning applications can run on those edge processors. Using local IoT networks, connecting several routers, gateways, and servers, a substantial part of the analytics and decision-making can happen on the edge networks.

As we deploy more edge computing-capable gateways, we need to be careful when transferring a significant amount of processing from the cloud layer to the edge layer, especially for CPU intensive applications. Heterogeneous data preprocessing and data training require considerable computing resources, but it is difficult for clients to complete these tasks because of their capability limitations.

That’s why companies such as Google and Huawei are already producing specific products to augment the IoT computing capabilities on the edge.

When the number of tasks is much bigger than the processing capability of the edge servers, we can avoid potential gateway failures by scheduling processing across the local network. Efficient scheduling is needed to optimize machine learning for IoT in the edge computing structure.

As task scheduling has little information about future tasks, the deployment decision initially relies on the historical tasks. Once machine learning algorithms can analyze the performance of the edge servers and gateways, and the efficiency of performing a large number of tasks, the system can start predicting the optimal scheduling across the network.

Another approach is using machine learning in a Transparent Computing (TC) environment, where training tasks move from lightweight clients to servers and edge devices. TL supports various machine learning algorithms, such as deep learning (DL) and support vector machine (SVM). The algorithm is pre-configured in the system, and the corresponding software packages installed before data training. TL can reduce the training time considerably on the premise that it ensures accuracy.

Once we start applying machine learning the edge network can perform many of the tasks usually reserved for cloud servers, and provide:

• Increased operational reliability: Using local storage and processing, deriving intelligence from local data, it is possible to deploy more efficient IoT solutions. This is especially important in locations where cloud connectivity is intermittent or suffers a high rate of latency.

• Real-time predictions: Running on-device machine learning models, an edge computing IoT network can achieve significantly faster predictions for critical IoT applications than typical cloud computing solutions, with the added advantage of ensuring data privacy.

• Increased security: As the edge network, paired with machine learning, can process and analyze data such as images, videos, sounds, and other sensor collected data locally on edge devices, it addresses specific privacy and compliance needs, thus reducing privacy and security risks.

BIG COMPUTER COMPANIES ARE JUMPING IN

Last month, during its Cloud Next ’18 conference, Google announced a couple of new products for machine learning and edge computing on IoT: Edge TPU, which, according to the company “enables the deployment of high-quality ML inference at the edge”, and Cloud IoT Edge, which extends Google Cloud’s powerful data processing and machine learning to billions of edge devices, such as robotic arms, wind turbines, and oil rigs, so they can act on the data from their sensors in real time and predict outcomes locally.

“Real-time decision-making in IoT systems is still challenging due to cost, form factor limitations, latency, power consumption, and other considerations. We want to change that,” says Injong Rhee, VP, IoT, Google Cloud. “We’re announcing two new products aimed at helping customers develop and deploy intelligent connected devices at scale: Edge TPU, a new hardware chip, and Cloud IoT Edge, a software stack that extends Google Cloud’s powerful AI capability to gateways and connected devices. This lets you build and train ML models in the cloud, then run those models on the Cloud IoT Edge device through the power of the Edge TPU hardware accelerator.”

To truly realize edge computing in IoT applications, there are still many challenges to be addressed, such as how to efficiently distribute the processing needs across devices, servers, and gateways. Solving this challenge and others remains the promise of Artificial Intelligence with its machine and deep learning abilities.

Machine learning can become a robust analytical tool for vast volumes of data. The combination of machine learning and edge computing can filter most of the noise collected by IoT devices and leave the relevant data to be analyzed by the edge and cloud analytic engines.