Beefy ML: Ultra-Low Power Algorithms on Cattle
summary
Machine learning runs in microseconds on a basic Cortex M4F, enabling an ear‑tag system that estimates grazing time and bite rate from accelerometer data. The stack uses Zephyr + Zephyr Power Management, training in PyTorch and embedded inference with CMSIS‑DSP; models are tiny (about 3 KB ROM) and run a 5 s window in ~20 ms on devices with 1 MB flash / 256 KB RAM. Over 100 multi‑month collar deployments validate the approach, but drinking often confuses grazing in the confusion matrices.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:57Why ML on cattle matters
- 9:54Hardware for monitoring
- 16:19Data processing pipeline
- 21:37Algorithm accuracy metrics
- 26:51Energy efficiency & acceleration
- 31:45Future directions
- 34:33Impact on farming & emissions
- 38:46Q&A
related talks
41mBeefy ML: Ultra-Low Power Algorithms on Cattle
25mMachine Learning on Microcontrollers; Tiny, Useful and Fun
31mKeynote: The next big thing in machine learning is tiny
45mUsing Edge Machine Learning to save power and development time
01h 53mTinyML: Machine Learning’s Future is Tiny and Bright
19mWhy Machine Learning on Microcontrollers?
26mNext generation machine learning for mobile and embedded platforms
27mWhat’s TinyML good for