What is edge AI?

Edge artificial intelligence (AI), or AI at the edge, is the use of AI in combination with edge computing to allow data to be collected at or near a physical location. For example, an image recognition algorithm task will run better being closer to the source of the data.

Edge AI allows responses to be delivered almost instantly. With edge AI, data is processed within milliseconds providing real-time feedback with or without internet connection because AI algorithms can process data closer to the location of the device. This process can be more secure when it comes to data because sensitive data never leaves the edge.

Edge AI differs from traditional AI because instead of running AI models at the backend of a cloud system, they run on connected devices operating at the network edge. This adds a layer of intelligence where the edge device not only collects metrics and analytics but is able to act upon them via an integrated machine learning (ML) model within the edge device.

The goal of artificial intelligence is the same – to have computers collect data, process that data, and then generate results similar to human intelligence. However, edge AI does the work and decision making locally, inside, or near whatever device being used. 

What is AI inference? 

How is edge AI different from distributed AI?

Related to edge AI is distributed AI which uses concepts from both traditional AI and edge AI. The main distinctions between the 2 appear where data is processed and how the AI models are deployed. Where edge AI runs algorithms directly on edge devices, distributed AI uses multiple interconnected systems–central servers, edge devices, and others.

In distributed AI, tasks are divided among several machines or devices, and each works on a portion of the problem. While this distributed processing can apply more power to the processing of data and is scalable beyond the capacity of edge AI, the trade-offs appear in its complexity, latency, and overall privacy.

How does cloud computing empower edge AI?

It is no exaggeration to claim that edge AI could not exist without cloud computing. Cloud computing provides the infrastructure, tools, and services to develop, deploy, manage, and maintain AI models on edge devices.

Training: Because edge AI devices exist away from centralized servers, they typically lack the computational power and large volumes of data necessary to train deep learning models. Edge AI devices instead transfer their data to a cloud where it is combined with that of similar devices, processed, and used to train the model. Trained machine learning models are then re-deployed to devices on the edge.

Deployment: Because edge devices are minimal by design, trained models need to be optimized for those resource-limited edge devices. Cloud services provide compression tools for quantizing and pruning that prepare AI models for deployment to the edge.

Data sync: Edge AI devices are able to process data quickly at their point of deployment. Edge AI devices also collect data in order to train their models to make better decisions. Edge AI devices regularly sync with a central repository in the cloud, which helps with storing and processing the data that the edge device is collecting and computing. Data sent to the cloud is used in continuous learning where the models are trained and re-deployed to the devices.

Monitoring and management: Edge AI devices are at the forefront of an organization’s interaction with their users. Cloud platforms monitor edge devices in real-time, enabling predictive maintenance and identifying potential issues before they impact performance. Additionally, cloud platforms can scale as needed offering elastic resources for an organization managing a fleet of devices.

Learn more about Red Hat AI