Season 7 of IoT Unplugged kicked off with an insightful conversation with Satyajit Sinha, Principal Analyst of IoT Analytics, as he tracked the evolution of Edge AI, from tiny ML models to multi-modal VLM, as well as discussed the different types of the Edge.
The emergence of Edge AI oversaw the transition of running models from the Cloud to the Edge, as the higher bandwidth, faster processing, and reduction in cost were all attractive benefits.
Sinha explained there are different kinds of Edge computing, including the micro Edge, thick Edge, and thin Edge. These types all dictated different chipset requirements and had different capabilities, which requires organisations to think about what kind of problem they are hoping to solve and what capabilities they need.
Some use cases are arguably more viable than others from a commercial perspective – Sinha mentioned machine vision as one example, but also the automotive industry, where growing numbers of connected vehicles need data processed quickly, with low bandwidth, and Edge AI can provide this. No one use case should be treated the same.
Sinha noted a trend in software becoming tightly integrated with hardware as a result of Edge AI applications growing in size and scale. This requires organisations to not only test AI models before deployment, but also to ensure these models are properly trained on the right data.
“This was very fragmented,” said Sinha. “A few companies that I’m going to talk about, quickly jumped on the problem and tried to solve them. Qualcomm is one of them. They created their own AI hub, where you have the opportunity to take the models and finetune them.”
Security risks should be brought into the conversation because although Edge AI has been lauded for being more secure compared with Cloud processing, there are questions that need to be asked: for example, when training models with data, to ensure they cannot be accessed by anyone else.
To hear what Sinha had to say about the evolution of Edge AI, software, and security, listen on Spotify, Apple Podcasts, and at the link below.
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by visiting our LinkedIn page.
