Edge AI and On-Device Machine Learning #1007459

Edge AI and On-Device Machine Learning is a practical engineering guide to running models where the data is generated — on phones, embedded devices, and in browsers. It starts from a single observation: the default location for inference is shifting from the cloud to the edge, and the teams that understand the constraints and trade-offs of edge deployment will build the products that win.

The book walks through the full stack — the case for edge AI and where it fits, hardware tiers from microcontrollers to mobile NPUs to browser GPUs, quantization and post-training compression, knowledge distillation for small students, pruning and structured compression, the mobile inference runtimes (TFLite, Core ML, ONNX Runtime, NCNN, MNN), browser-based ML with WebAssembly and WebGPU, TinyML on microcontrollers, federated learning and differential privacy, on-device training and adaptation, deployment and update mechanisms, evaluation methodology for edge systems, security and adversarial robustness, and the trends reshaping the field.

It covers the failure modes that quietly wreck edge deployments: a model that runs at 30 FPS on a flagship phone and 2 FPS on a mid-range device, a quantization step that reduces accuracy by fifteen points on the tail, a runtime that selects a slow kernel because hardware capabilities were misdetected, an update that exceeds the device's storage budget and fails silently, a TinyML model that fits in flash but not in RAM at inference time, an NPU that falls back to CPU and adds seconds to every inference. Each is presented with the failure, the countermeasure, and the operational tradeoff.

Fourteen chapters, ~50,000 words. Real Python code with TensorFlow Lite, Core ML, ONNX Runtime, TensorFlow.js, TensorFlow Model Optimization, and TinyML. Written for engineers who need models to work on real devices, not just in a notebook.