Edge AI

Edge AI studies how to run and learn with AI close to where data is produced, from microcontrollers and wearables to robots, industrial systems, and nearby edge servers. Common approaches include quantization, pruning, and distillation; hardware-aware runtimes for CPUs, GPUs, and NPUs; and split inference or federated learning across devices and cloud services. Evaluation must balance task quality with latency, energy use, memory, and reliability on real hardware, often using vision, speech, sensor, and language tasks. A central challenge is making systems adapt to changing workloads and conditions while remaining safe and dependable across heterogeneous devices, without assuming that compression or simulated performance will translate to field results.