2021 saw massive growth in the demand for edge computing - driven by the pandemic, the need for more efficient business processes, as well as key advances in the Internet of Things, 5G and AI.In a study published by IBM in May, for example, 94 percent of surveyed executives said their organizations will implement edge computing in the next five years.
From smart hospitals and cities to cashierless shops to self-driving cars, edge AI - the combination of edge computing and AI - is needed more than ever.
Businesses have been slammed by logistical problems, worker shortages, inflation and uncertainty caused by the ongoing pandemic. Edge AI solutions can be used as a bridge between humans and machines, enabling improved forecasting, worker allocation, product design and logistics.
Here are the top five edge AI trends NVIDIA expects to see in 2022:
1. Edge Management Becomes an IT Focus While edge computing is rapidly becoming a must-have for many businesses, deployments remain in the early stages.
To move to production, edge AI management will become the responsibility of IT departments. In a recent report, Gartner wrote, Edge solutions have historically been managed by the line of business, but the responsibility is shifting to IT, and organizations are utilizing IT resources to optimize cost. 1
To address the edge computing challenges related to manageability, security and scale, IT departments will turn to cloud-native technology. Kubernetes, a platform for containerized microservices, has emerged as the leading tool for managing edge AI applications on a massive scale.
Customers with IT departments that already use Kubernetes in the cloud can transfer their experience to build their own cloud-native management solutions for the edge. More will look to purchase third-party offerings such as Red Hat OpenShift, VMware Tanzu, Wind River Cloud Platform and NVIDIA Fleet Command.
2. Expansion of AI Use Cases at the Edge Computer vision has dominated AI deployments at the edge. Image recognition led the way in AI training, resulting in a robust ecosystem of computer vision applications.
NVIDIA Metropolis, an application framework and set of developer tools that helps create computer vision AI applications, has grown its partner network 100-fold since 2017 to now include 1,000+ members.
Many companies are deploying or purchasing computer vision applications. Such companies at the forefront of computer vision will start to look to multimodal solutions.
Multimodal AI brings in different data sources to create more intelligent applications that can respond to what they see, hear and otherwise sense. These complex AI use cases employ skills like natural language understanding, conversational AI, pose estimation, inspection and visualization.
Combined with data storage, processing technologies, and input/output or sensor capabilities, multimodal AI can yield real-time performance at the edge for an expansion of use cases in robotics, healthcare, hyper-personalized advertising, cashierless shopping, concierge experiences and more.
Imagine shopping with a virtual assistant. With traditional AI, an avatar might see what you pick up off a shelf, and a speech assistant might hear what you order.
By combining both data sources, a multimodal AI-based avatar can hear your order, provide a response, see your reaction, and provide further responses based on it. This complementary information allows the AI to deliver a better, more interactive customer experience.
To see an example of this in action, check out Project Tokkio:
3. Convergence of AI and Industrial IoT Solutions The intelligent factory is another space being driven by new edge AI applications. According to the same Gartner report, By 2027, machine learning in the form of deep learning will be included in over 65 percent of edge use cases, up from less than 10 percent in 2021.
Factories can add AI applications onto cameras and other sensors for inspection and predictive maintenance. However, detection is just step one. Once an issue is detected, action must be taken.
AI applications are able to detect an anomaly or defect and then alert a human to intervene. But for safety applications and other use cases when instant action is required, real-time responses are made possible by connecting the AI inference application with the IoT platforms that manage the assembly lines, robotic arms or pick-and-place machines.
Integration between such applications relies on custom development work. Hence, expect more partnerships between AI and traditional IoT management platforms that simplify the adoption of edge AI in industrial environments.
4. Growth in Enterprise Adoption of AI-on-5G AI-on-5G combined computing infrastructure provides a high-performance and secure connectivity fabric to integrate sensors, computing platforms and AI applications - whether in the field, on premises or in the cloud.
Key benefits include ultra-low latency in non-wired environments, guaranteed quality-of-service and improved security.
AI-on-5G will unlock new edge AI use cases:
Industry 4.0: Plant automation, factory robots, monitoring and inspection.
Automotive systems: Toll road and vehicle telemetry applications.
Smart spaces: Retail, smart city and supply chain applications.
One of the world's first full stack AI-on-5G platforms, Mavenir Edge AI, was introduced in November. Next year, expect to see additional full-stack solutions that provide the performance, management and scale of enterprise 5G environments.
5. AI Lifecycle Management From Cloud to Edge For organizations deploying edge AI, MLOps will become key to helping drive the flow of data to and from the edge. Ingesting new, interesting data or insights from the edge, retraining models, testing applications and then redeploying those to the edge improves model accuracy and results.
Wi
