Intermittent Learning: On-Device Machine Learning on Intermittently Powered System
summary
Framework schedules ML subtasks around harvest interrupts and uses an algorithm plus three runtime heuristics to decide which examples to keep or discard, reducing wasted training work. Implements split-up k-NN, k-means and a small neural net so learning resumes across power failures. Built three battery-less apps: air quality (solar), human presence (RF) and vibration (piezoelectric); on-device training reaches 81%, 82% and 76% accuracy and reports up to 100% energy-efficiency gain while cutting examples by up to 50%.
AI-written summary · about
chapters (7)
- 0:00Introduction to the concept
- 0:58Challenges in the field
- 2:17System execution process
- 2:42Proposed framework modules
- 4:28Real-world applications
- 5:02Air quality monitoring case
- 6:00Conclusion
related talks
57mMachine Learning in the Age of IoT
25mMachine Learning on Microcontrollers; Tiny, Useful and Fun
31mKeynote: The next big thing in machine learning is tiny
10mMachine Learning for mobile
45mUsing Edge Machine Learning to save power and development time
39mMachine Learning on Android Demystified
15mSimplified Machine Learning on mobile, browsers, and edge devices with on-device ML
30mDeep Learning in Energy Management: Non-Intrusive Load Monitoring for IoT Devices