Machine learning (TinyML) classifier that trains on the microcontroller itself, in one pass over the data.
SEFR (arXiv:2006.04620) brings on-device learning and AI to small boards: learn a multiclass linear classifier from sensor samples streamed one at a time, with no learning rate, no iterations and no dynamic memory. An optional second pass (LDA) improves multiclass accuracy. Runs on 2 KB RAM boards such as the Uno, and on ESP32, RP2040 and nRF52. The SEFR algorithm is also used in STMicroelectronics NanoEdge AI Studio.
| Filename | Release Date | File Size |
|---|---|---|
| SEFR-1.0.1.zip | 2026-10-19 | 15.74 KiB |
| SEFR-1.0.0.zip | 2026-10-19 | 14.87 KiB |