Explainable AI User Study
Understanding how AI models come to their decisions is a matter of increasing importance – for finding systematic errors, exposing bias, and convincing EU juries (explanations being guarenteed under the GDPR). Over 10 weeks, I embarked on a research project with a group of peers to understand three explainable AI models; both how they worked, and how convincing their explanations were to others.
Content-recognizing Compost Bin
I wired together a Raspberry Pi with a camera, light, and accelerometer to create a bin that would monitor its contents and automatically upload pictures to Dropbox. Ultimately, these pictures will train a neural network classifier running on-device to determine whether the contents are compostable.
Asshole-predicting Neural Network
As an extension of previous social computing work, a friend and I created a word vector model and a neural network to classify posts on Reddit's "am I the asshole" subreddit. We decided to use the valence of each word to classify posts as asshole or not, in the vein of O'Brien, 2020.
Dark Souls Armor Optimizer
Playing through Dark Souls 1, I was struck by the difficulty of choosing what to wear. With dozens of armor pieces for each body part, and weight limits/elemental defenses/poise/etc to juggle, I decided it would be easier to make a publicly-available program to decide between all options.