Machine Learning

ML music videos, Presidents in Picasso, Sound-controlled Zelda, and other AI weirdness.

ROLE: machine learning, facial recognition, protocols

TECH: tensorflow,paperspace,ml5,style transfer, word vectors, openFrameworks, javascript, C++

What happens to the soul?

I taught a workshop in which students made a Machine Learning generated music video using BigGan, Magenta, and lyrics from an AI trained on Ludacris

Proximity driven text size adjuster

Webcam Proximity Text Size Adjuster

Using Machine Learning to adjust the size and characteristics of variable fonts according to a user's proximity to the screen.

Word Vector Live Thesaurus

Real-Time Word Suggester

Suggests word embedding-based suggestions for text as you type.

Playing Games with Sound

In this video I've trained a model to distinguish between various sounds - whistling, drumming, clapping, and my best lil' jon impersonation. Each sound controls a key on the keyboard. In this video I'm using sound to control a game of Zelda.

Arbitrary Style Transfer with No Model Training Needed

Arbitrary Style Transfer

Normally, training a model on a specific style takes many hours of intensive GPU processing, and no small amount of $money. Here is some code that uses transfer learning to get around all that.

Machine learning style transfer using ml5 and tensorflow

Style Transfer

Training a model on the style of one image in order to render that style to any random image - a sort of hyper-advanced photoshop filter. This example uses Hokusai's Great Wave off Kanagawa