I was the chief technical officer at the Machine Learning and Data Science (MLDS ) club at UT Austin for 5 years.
My responsibilities included:
Writing up meeting workshops on Machine learning/Data science concepts
Designing/Testing Semester Competition Datasets
Designing/Testing Hackathon Datasets
I particularly enjoy designing, building, and running the semester competitions and hackathons the club puts on. See below for some of the work I've done building these competitions over the past few years.
Built all Modeling, Texturing, Lighting and Programming/Rule Generation and Competition Design.
In this competition, students were tasked with identifying which objects were present in a room scene of synthetically generated objects. Using python I built a system within to randomly place ~50 different household objects within a room under realistic constraints, then render them with Blender. Students were given 10,000 room images for training and made predictions for the remaining 10,000. Read more about the competition here. To see the code/blender file for creation of the competition dataset, see my github here.Â
Co-Built and Tested all Code and Competition Design.
For this competition students were tasked with creating an AI to play a modified, competitive, game of life. Each square could attack all squares within a 5*5 neighborhood, with attacked cells coming under control of the team they were attacked by. Like game of life, if cells were surrounded by too many active cells, they would die. In teams of 2-4, students would submit a script that defined their agent's actions, and team scripts were then pitted against each other in a live competitive bracket. Setting up this competition both years was a joint effort between myself and Henry Castillo, who now works at Stripe. You can take a look at some example code, and even run a match yourself, here.
Designed Competition and Generation System
Within this competition for 2020, students were asked to identify handwritten characters in noisy 20*20 images. For this competition I wanted to avoid students using off the shelf models so all characters were custom made and animated using scripts in blender. To keep word and letter frequency reasonable, characters were converted over from "To Kill a Mocking Bird", leaving a second, cryptographic puzzle for students who developed an accurate enough classifier.