True TinyML with Weights & Biases - Building a Sound Classifier

About the Event

While many of us are familiar with TinyML by way of devices like Siri or Alexa, TinyML has recently taken off in the open source community with democratically accessible devices such as the Raspberry Pi Pico and similar microcontrollers. True TinyML applications run on microcontrollers with only several hundred KB of RAM without operating systems and can perform continuous inference on extremely low amounts of power. TinyML requires: 1) Models that only use micro watts of power of inference 2) Devices have limited RAM 3) Devices are physically tiny TinyML is in the realm of switch-type tasks that initiate more complex, sophisticated, and resource-intensive tasks, where constant listening/real-time inference takes place. In this webinar, we'll demonstrate how to train and deploy a model from scratch using Tensorflow Light, then evaluate and tune a TinyML model using W&B. This is meant to be a step-by-step walkthrough that ends with a reproducible example for all attendees. In this webinar, we covered the following: 1) Data storage, acquisition, and versioning using Artifacts 2) Pre-processing and data visuality using W&B Tables 3) Hyperparameter tuning using Sweeps 4) Selecting and storing models using Model Registry 5) Deployment/Inference monitoring