Lecture
From cloud-first to edge-native: rethinking AI for the next billion devices
- at -
- Hall C6C6.501
- Language: English
- Type: Lecture
Lecture description
While the recent past has seen big moves toward cloud technology, we’re now entering an age of Edge AI, where models run on microcontrollers at the edge of a network, close to the source of data. There are two ways to approach this move. One option is to focus entirely on deployment, and try to get models designed for the cloud to run on the edge. The other option is to take an Edge-native approach for a more intentional design; maximizing efficiency, adaptability, and trust. This talk will show the benefits of taking the Edge-native approach, and the potential pitfalls of focusing solely on deployment. Specifically, it will focus on the areas of opportunity (or potential problems) that need to be considered when making this choice: the abundance of model zoos filled with imperfect models, security risks, and memory and power constraints. The talk will also explore the perspectives and unique challenges that machine learning (ML) engineers and embedded engineers each face in the world of Edge AI, and what they each face when getting AI models to run successfully on Edge devices. This presentation is intended for ML and embedded practitioners in any industry. It is especially relevant for anyone who has faced the challenges involved in bringing ML applications to production on Edge devices. Attendees will leave with a clearer picture of why and how we need to design AI applications that are truly Edge‑native, and an understanding of how ML and embedded teams can collaborate more effectively.
