Key trends in AI inference for 2022 - Embedded.com

Key trends in AI inference for 2022

The ability to perform AI inferencing closer to the end user is opening up a whole new world of markets and applications.

It’s an exciting time to be a part of the rapidly growing AI industry, particularly in the field of inference. Once relegated simply to high-end and outrageously expensive computing systems, AI inference has been marching towards the edge at super-fast speeds. Today, customers in a wide range of industries – from medical, industrial, robotics, security, retail and imaging – are either evaluating or actually designing AI inference capabilities into their products and applications.


Dana McCarty (Source: Flex Logix)

Fortunately, with the advent of new semiconductor devices developed specifically to accelerate AI workloads, this technology has now advanced to the point where many products have dropped to price points and form factors that make it viable for mainstream markets where AI can be incorporated into a wide range of systems.

As we look to 2022, here are our predicted AI inference trends.

Security, Privacy Concerns

We will continue to see growing privacy concerns as AI is deployed more broadly. Techniques that obscure or protect personal details will expand, as will techniques to secure AI systems. Among them will be the application of root-of-trust technology against cyber-intrusion.

Continued Model Evolution

The industry will shift from models developed five to seven years ago such as MobileNet and ResNet toward new, more powerful and accurate approaches like Yolo-v5 and transformer-based solutions. Continuing research into AI inference models seeks to provide greater accuracy and higher performance. Deployed systems must be designed so that the models can be updated over time to improve their performance and accuracy as new techniques are discovered.

Edge Migration

Edge transition will continue as companies scale applications; economics will push them to offload bandwidth and compute-heavy applications such as computer vision from the cloud-to-edge devices. Customers will increasingly adopt AI acceleration where high accuracy, high throughput and low power on complex models is needed.  For example, in the industrial segment, AI could be used to help manage inventories, detect defects or even predict defects before they happen.

We expect this technology to also expand into many other edge applications such as surveillance, facial recognition, gene sequencing, medical imaging and more.

Open Source

Open source will continue to be the primary platform for AI development, with Python-based tools gaining ground. While there is a clear desire to ensure that model frameworks remain open, training data used by actual systems and the driver of accurate models is more likely to be proprietary and therefore closely guarded by AI developers.

Next Unicorns

Edge AI silicon providers will emerge as the next set of unicorns as AI expands to the edge. As this technology is now available across a wide range of industries, more companies will want to leverage its capabilities to deliver new innovations or to differentiate themselves from competitors.

The ability to perform AI inferencing closer to the end user is opening up a whole new world of markets and applications. It will be exciting to see new products entering the market in 2022.

–Dana McCarty is vice president of sales and marketing for inference products at Flex Logic Technologies.

>> This article was originally published on our sister site, EE Times.


Related Contents:

For more Embedded, subscribe to Embedded’s weekly email newsletter.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.