interviewed at the Artificial Intelligence Conference, London 2018 – Pete Warden
summary
Cheap STM32 "Blue Pill" boards are programmed bare‑metal (OpenOCD via Raspberry Pi) using code on stm32_bare_lib GitHub and run on 20 kilobytes RAM, 64K flash hardware bought for <$2. A prototype fits a 20,000-parameter 8-bit conv net (about 20 KB in flash) that classifies a 1‑second spectrogram built from 20 ms FFT frames, runs in a few tens of kilobytes of RAM, and targets ultra‑low‑power keyword and anomaly detection on coin‑cell batteries.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:25Whats this thing
- 5:05Applications
- 6:40Low power network
- 7:35Machine learning
- 9:20Mobile devices
- 11:00Deep learning models
- 12:25Image and audio recognition
- 14:50Listening and predictive maintenance
- 16:50Environment and adaptation
- 19:20How are you accomplishing this
- 21:15What is Bare Metal
- 22:35Tensorflow
- 24:25Looking ahead
- 26:40Edge and cloud
related talks
24mRunning AI/Neural networks on microcontrollers made simple with the STM32Cube.AI
33mmruby on the minimal embedded resource
31mKeynote: The next big thing in machine learning is tiny
31mReally Small Rust
06mEmbedded Executive LIVE
11mLightning Talk: Accelerating Experimentation...
49m17,000 contributions in 32K RAM
31mRunning Linux on Constrained IOT Device