Deep Learning in the Wild: Embedded PyTorch for...
summary
Owl IoT “ducks” form a LoRa mesh that hops detections to a single gateway and then to satellite; initial deployment runs BirdNet on Raspberry Pi, gets detections but suffers false positives and three‑day battery life. Phase two adds DSP preprocessing that extracts ~20 features, swaps to a Teensy 4.1, evaluates random forests and tiny neural nets with inference under 1s, and ships a 100 kilobyte model that fits power, memory, latency, and imbalance constraints.
AI-written summary · about
chapters (9)
- 0:00Embedded deep learning project
- 1:30Conservation challenge: bird species
- 2:55Three-body problem in edge AI
- 3:56Connectivity and model deployment
- 5:41Power and model improvements
- 6:23Digital signal processing
- 7:48Model performance and hardware
- 8:31Summary and next steps
- 9:36Key takeaways
related talks
23mBirding with Python and Machine Learning
28mChasing birds with code
57mMachine Learning in the Age of IoT
35mAI is for the birds
59mML and the IoT: Living on the Edge
43mBuilding the Duckling Droid: AI, Robotics, and Computer Vision in Action
31mKeynote: The next big thing in machine learning is tiny
43mKeynote: Tiny Data Collectors, Vastly (...) & the Land of Tiny Challenges