Students worked with state-of-the-art tools to uncover patterns in a never-before analyzed neural data set at Brown’s first-ever Neurodatathon Nov. 28 to Dec. 2.
Hosted by Brown Data Science and sponsored by the Brown Institute for Brain Science and the Brown Initiative for Computation in Brain and Mind, the Neurodatathon challenged students to analyze a neural data set with a variety of machine learning and deep learning methods.
“What we were hoping, by introducing this brand new data set, was that students would have an opportunity to figure out interesting characteristics about the data and also (discover) optimal ways of filtering it, so that optimal results can be obtained when doing scientific analysis,” said Drew Linsley, postdoctoral research associate in cognitive, linguistic and psychological sciences.
“We had 140 students sign up, who had concentrations ranging from physics and computer science to neuroscience and psychology,” Linsley said.
With students coming from such wide-ranging fields, the event featured two tracks of competition, each demanding a different set of skills and interests. Participants either formed teams or worked independently, competing in one of the two tracks.
The “Optimize” track tended to be the choice for programmers and deep learning and machine learning enthusiasts. In this track, the goal was to build a model that most accurately interpreted neural behavior, as given by the data set. The two groups whose models generated the most accurate results won the competition.
The “Create” track involved scientific analysis and was chosen more often by students with stronger neuroscience backgrounds. Because the provided data set had never been analyzed before, participants were given lots of freedom and an opportunity to come up with scientifically significant analysis. Three neuroscience and data science experts judged the analysis submitted by the participants, and the two winners were chosen for the impact and creativity of their submissions.
“In a perfect world, you would be an expert in both (computing and neuroscience), but of course it’s very difficult to get this kind of training, especially at the undergraduate level,” Linsley said. “Our special two-track competition catered to the two different types of students interested — the programmers who had more technical skill and the neuroscientists who had more knowledge about the brain.”
Because many of the participants had limited or no experience with machine learning algorithms, neural data sets and even data science in general, the organizers of the event hosted workshops. In addition, office hours were held every day during the event, giving participants opportunities to talk to deep learning and neuroscience experts.
A hacking space was offered for participants at night, giving students a place to work on projects and providing them access to experts. The workshops prior to the Neurodatathon “made the event much less daunting,” said Dylan Sam ’21. “It was a little bit of an introduction and walkthrough, which was really great. The experts did a really good job explaining the concepts to (participants).”
Mariel Rosic ’20 agreed that the workshop was a helpful aspect of the event. “The postdocs and grad students were so helpful,” she said. “I learned so much by talking to them, and it almost felt like a series of one-on-one machine learning sessions that helped me with my analysis.”
Participants and event hosts agreed that the event set a positive precedent for future Neurodatathons.
“We ended up finding some really interesting analysis, discovering interesting relationships between brain and behavior that we didn’t expect we would find,” Linsley said. The Brown Data Science Initiative will continue hosting events next semester, with plans of coordinating another Neurodatathon in the near future, he added.
