TinyML as-a-Service
summary
Proposal of TinyML-as-a-Service delivers an end-to-end pipeline that trains in the cloud, compiles models to run on Cortex‑M class microcontrollers, and deploys them with partial over‑the‑air updates. It uses three enablers: Unikernel (MirageOS + OCaml) for tiny runtime images, an ML compiler stage (supports multiple backends including TensorFlow Lite for Micro), and CoAP/LwM2M orchestration. A working demo compiles and installs models into megabyte‑class devices and performs OTA module installs successfully.
AI-written summary · about
chapters (8)
- 0:00Introduction to TinyML
- 0:47What is TinyML
- 3:35Local processing vs cloud upload
- 7:40Customizing models for devices
- 9:04TinyML as-a-Service workflow
- 12:46Building the TinyML ecosystem
- 20:31Call for ecosystem partners
- 23:29Q&A
related talks
01h 53mTinyML: Machine Learning’s Future is Tiny and Bright
27mWhat’s TinyML good for
36mTinyML at the Edge: Deploying and Optimizing AI Workloads on Zephyr RTOS
18mManagement of IoT TinyML Devices
17mAI workflow for large scale deployment of far-edge ML devices
25mMachine Learning on Microcontrollers; Tiny, Useful and Fun
01h 01mML and the IoT: Living on the Edge
10mMachine Learning for mobile