Designing Efficient Neural Architectures and Scaling Strategies for Edge Computing

Learn about efficient neural architectures and scaling strategies for edge computing in this technical talk presented by Francesco Paissan, a Junior Researcher from the Energy Efficient Embedded Digital Architectures Unit at Fondazione Bruno Kessler. Explore how distributing intelligence from cloud to edge computing becomes crucial for infrastructure sustainability in the IoT era, offering advantages in power efficiency, low-latency inference, and privacy-by-design. Discover novel approaches to overcome tinyML challenges such as limited memory and energy constraints through innovative neural architectures, training paradigms, and scaling strategies. Examine efficient multimedia analytics pipelines that deliver state-of-the-art results while using significantly fewer computational resources than traditional methods. Gain practical insights into the application of these methodologies in object detection, tracking, and zero-shot audio classification use cases.