Edge computing at the edge of the world
summary
Builds a real-time music recommender that runs on Adafruit ItsyBitsy microcontrollers using CircuitPython, extracts song features by converting tracks to mel-spectrograms and running VGG16 on mel-spectrograms (Keras) to get 1,000‑dim vectors, clusters them with agglomerative (Ward, Euclidean) into folders, and maps short-term flight telemetry to folders with a lightweight Naive Bayes + multi-armed bandit (ε‑greedy / UCB ideas). Learns a folder in about 5–8 songs in simulator, deploys on the plane with working telemetry and SD‑card playback.
AI-written summary · about
chapters (10)
- 0:00Introduction and title joke
- 1:29What is a makerspace
- 3:57Circuit Python overview
- 9:29Circuit Python demo
- 10:10Automatic music categorization
- 19:23Reinforcement learning setup
- 23:19Multi-armed bandit explained
- 30:22Final project and demo
- 31:46Takeaways and resources
- 32:33Q&A session
related talks
27mMicrocontrollers and Machine Learning with MicroPython and emlearn
45mAI Lofi HipHop - How Not To Generate AI-Music
38mmusic2vec: A smart music recommendation and discovery
54mSmarter Everything with ML & the IoT
41mCreating Intelligent Music Instruments with Machine Learning
43mKeynote: Tiny Data Collectors, Vastly (...) & the Land of Tiny Challenges
22mAI On The Farthest Edge
31mHaving fun with Music and Keras