Edge AI

Running models where the data is produced: microcontrollers, gateways and phones. Covers compression and quantization, IoT system design and Android apps with on-device models in android/.

This part builds on Machine learning and Neural networks.

0 of 17 topics ready, listed in reading order

The edge AI landscapeCloud, edge and on-device inference, the runtimes available and how to choose.Planned

TinyML and quantizationMicrocontroller constraints and integer quantization with measured accuracy and size.Planned

Model compressionPruning, distillation and efficient architectures.Planned

Edge anomaly detectionA sensor autoencoder exported to an int8 runtime.Planned

Edge, fog and cloudDeciding where each computation happens, with a tiered simulator.Planned

IoT messagingHTTP, MQTT and CoAP compared on real traffic.Planned

Device connectivityBLE, Zigbee, LPWAN and cellular IoT and what each allows.Planned

Energy-aware sensingDuty cycles, energy budgets and sending less data.Planned

Digital twins and predictive maintenanceA live twin over MQTT and remaining-useful-life models.Planned

Activity recognitionClassifying human activity from wearable sensors.Planned

Intrusion detectionDetecting attacks in device traffic under heavy class imbalance.Planned

Camera inference on AndroidA real-time camera pipeline with correct coordinate transforms.Planned

On-device language modelsRunning a small language model on a phone and measuring it.Planned

Language model clients on mobileA provider-agnostic streaming client with cancellation.Planned

Vision APIs on mobileText recognition, barcodes, faces and pose on device.Planned

Model deliveryBundling, downloading and verifying models on devices.Planned

Securing AI keysWhy keys in apps leak and what to do instead.Planned