Scalable Edge AI Architectures for Real-Time Compute Pipelines

1. Low-Power Edge AI: TinyML and Embedded Intelligence
The first part of the talk focuses on low-power Edge AI platforms designed for battery-powered and resource-constrained devices such as wearables, smart home products, and consumer IoT. It covers TinyML workflows using Edge Impulse and microcontroller-based inference.
2. High-Performance Edge AI: Accelerated Vision and Analytics
The second part examines high-performance Edge AI systems for compute-intensive applications including intelligent surveillance, and industrial automation. It explores heterogeneous computing architectures combining CPUs, GPUs, and FPGAs, with frameworks such as NVIDIA Holoscan enabling high-throughput streaming AI pipelines capable of processing multiple high-resolution sensor streams in soft real time.
3. Low-Latency Edge AI: Deterministic Sensor-to-Decision Systems
The final part discusses low-latency Edge AI architectures for mission-critical applications in defence, medical devices, robotics, and automotive systems, where deterministic performance is essential. Topics include FPGA-based designs, ASICS and heterogeneous semiconductor roadmaps from NVIDIA, AMD, Altera, and others, highlighting the industry's trajectory.