Kyle Phillips
Engineer & Creative · Google NYC
Tensorflow.js
Google's open-source library for running and training machine learning models in the browser.
Tensorflow.js is Google's open-source library for running and training machine learning models in the browser and on Node.js — no install, no server round-trip, just JavaScript.
It runs through several pieces of my work. Teachable Machine uses Tensorflow.js for in-browser transfer learning, letting anyone train a model from a webcam without writing code or sending data to a server. Creatability uses PoseNet — a Tensorflow.js model — to power the body-controlled interfaces I built with the disability community.
The browser as an ML runtime mattered to both projects: it kept the data on-device, removed the friction of installing anything, and made the work accessible to anyone with a webcam.
2 References
One of my central contributions was an open-sourced library of Web Components solving many complex accessibility issues and providing novel modes of interaction such as using PoseNet with Tensorflow.js to enable anyone to control a cursor with their nose, wrist or many other body parts with nothing other than a basic webcam.
Millions of models have been trained with Teachable Machine using transfer learning with Tensorflow.js; it remains the easiest way to introduce anyone to this technology that is shaping all of our lives. It is beyond rewarding when I am shown something created by others or am told their first introduction to AI was with Teachable Machine. There are so many projects I wish I could highlight.