This past weekend five of our team member’s headed to Boston for MIT’s annual student hackathon. Over the course of two days, nearly 1500 students from around the world came to MIT and build everything from an app to connect people in need of disaster relief and aid volunteers to a remote controlled robot butler.
The energy was high with students competing for prizes in multiple categories, including a prize for best Spell hack and best overall hack.
Since Spell is application agnostic, we decided to keep the Spell prize wide open. We wanted to give students free reign to be creative with Spell.
Our offer to hackers:We want to see your creative applications of deep neural networks. Maybe it’s an art project, a generated book of poetry, or an app that predicts housing prices. As long a it uses Spell’s cloud GPU, it’s fair game!
We had a number of amazing entries for our hackathon prize. The winners we picked for our first prize and runner up went on to win first prize and runner up for the overall HackMIT prize, so we know our prizes were well calibrated!
First place: YeetView
We gave our top spot to YeetView. Built by Mitchell McDermott and Rob Kopel from the University of Queensland, Australia, YeetView is a “fully accessible interface for Google StreetView.”
As users move around, YeetView will describe to them features within the space to help them navigate through voice.
The team trained “a custom CNN with a [sic] attention based LSTM trained on coco2014” on Spell to get text based descriptions of objects within a scene, and then used the Speech Synthesis API to “take a description string and read it back to the user.”
YeetView also got runner up for the overall HackMIT prize.
Runner up: ENNUI
ENNUI is a graphical interface where users can “construct any (non-recurrent) architecture they please.”
Built by Jesse Michel and Rikhav Shah of MIT, ENNUI ended up winning the overall MITHack prize.
Great for those just getting started with machine learning, ENNUI allows users to visualize their neural network architectures in a simple and intuitive way.
Once the user finishes building their neural net, ENNUI translates the diagram into runnable Keras code.
A weekend well spent
HackMIT was Spell’s first hackathon and it was a big success. It was inspiring to see students making insightful, funny, and technically ambitious projects in just a short weekend.
Until the next hackathon!
More on the winners
Rikhav Shah is a senior at the Massachusetts Institute of Technology majoring in pure math. He participates in a variety of campus activities, including the MIT Hackathon, where he led a team to win the Grand Prize two years in a row. Rikhav is also co-founder and CTO of Parsegon, a startup focused on digital math education. He also volunteers his time teaching fun topics in math to high school students.
I’m a senior at MIT studying math theory and computer science. I’m interested in understanding and proving properties of neural networks as well as more general problems in analysis and algorithms. Outside of the classroom you can find me contemporary dancing, distance running, or playing soccer.
Source: Deep Learning on Medium