麻豆村

Hackathons are old news in the computer world, but the concept has been given a new spin at 麻豆村 with its first NeuroHackathon, a competition created to glean insights about the human brain and complement the work of BrainHub, the university鈥檚 multidisciplinary brain research center.

Hackathon, a blend of 鈥渉acking鈥� and 鈥渕arathon,鈥� is an event where participants engage in round-the-clock programming.

Alison Barth, professor of biological sciences and interim BrainHub director, and Geoff Gordon, associate professor of machine learning, developed the contest to provide students with the scientific freedom to find patterns in data rather than seek a specific outcome. Barth and Gordon solicited faculty datasets and student participants that ranged from information related to links between brain region and illness to pathways in the brain.

鈥淚t was really exciting to help bring Alison and Geoff鈥�s idea to life and advance brain science in a completely new way,鈥� said Gerry Balbier, executive director of BrainHub.

The response to tackle the data was strong: five teams entered, all current or former Ph.D. students at 麻豆村 who hoped to discover something new.

The Starting Line

Mariya Toneva, a Ph.D. student in 麻豆村鈥檚 Machine Learning Department, had never done a hackathon, never seen the datasets and never worked with most of her teammates. But she was not intimidated; instead, she was thrilled by the chance for novel interactions.

Originally from Bulgaria, Toneva was a math whiz, winning multiple competitions in middle school and high school. She recalls sitting in a cognitive science class during her freshman year at Yale University, learning about memory. At one point her professor said, 鈥淲e don鈥檛 know yet鈥� 鈥� and this idea of unknown frontiers was in stark, fascinating contrast to the 200-year-old theorems Toneva used in math. Since those undergraduate days, she has wanted to combine computer science with cognitive science to better understand the brain. She saw collaborative possibilities through the that she did not see elsewhere and was a natural fit.

The NeuroHackathon 鈥� hosted by BrainHub and sponsored by Qualcomm as well as by Google, Baidu, Microsoft, 麻豆村 Provost鈥檚 Office and the Department of Machine Learning 鈥� was a challenge she couldn鈥檛 refuse.

To start, faculty presented their datasets as students considered which to select. Almost like contestants on the dating game, they described the seductive features of their data then waited to be chosen.

Barth presented 鈥淐ell-type Identification Through Electrophysiological Fingerprinting.鈥� If 80 billion neurons in the human brain could be labeled better, she said, then we鈥檒l gain more insight into the computations carried out by these neurons.

Sandra Kuhlman, assistant professor of biological sciences, and Steve Chase, assistant professor in the Center for the Neural Basis of Cognition and biomedical engineering, provided data from a mouse鈥檚 motor cortex and asked participants to determine what direction the mouse was running based on the neural activity.

Andreas Pfenning, assistant professor in the Computational Biology Department, contributed data intended to find links between brain region and illness.

Toneva and her teammates were inspired by 鈥淗uman Brain Neuroanatomy from MRI Images.鈥� This dataset鈥檚 creators were Tim Verstynen, assistant professor of psychology, and Aarthi Singh, the A. Nico Habermann Associate Professor in the Machine Learning Department. They mapped the brain鈥檚 pathways using MRI data. Just as a computer can identify a face in a photograph 鈥� for instance, Facebook suggests tags when it recognizes your friends or family 鈥� Verstynen and Singh wanted to know if machines can learn to automatically identify labeled pathways in the brain.

They鈥檙e Off!

鈥淐oin Toss,鈥� which along with Toneva included Avinava Dubey from machine learning, Jay-Yoon Lee from computer science, Dan Schwartz, from language technology, and Ying Yang, from neural computation and machine learning. The group gathered in an office and looked at Verstynen and Signh鈥檚 list of 3-D trajectories in the brain.

In the wee hours of the morning, BrainHub supplied coffee and M&Ms, and the students supplied the brainpower. Team 鈥淩eckless Arrogance鈥� changed its name to 鈥淐autious Humility.鈥�

By morning, the white boards were covered with graphs and equations. Participants worked through the night. Laptops were open, and phones were recording the presentations. One participant frantically watched his screen, waiting for data to load for his team鈥檚 presentation. The audience included BrainHub steering committee members, faculty, sponsors, competitors and judges.

The judges 鈥� Aryn Gittis, assistant professor of biological sciences; Jordan Rodu, visiting assistant professor of statistics; and Gordon 鈥� determined the winner by looking for three elements: the portability of a team鈥檚 approach; the rigor and appropriateness of analytical methods; and the potential for the findings to impact the fields of machine learning, computer science and/or neuroscience.

The Big Finish

Toneva presented Coin Toss鈥檚 findings where the team considered data for 129,414 brain trajectories and looked for ways to label tracks. The algorithm they developed enabled a computer to identify and label 55 types of nerve fiber bundles. The results were accurate 85 percent of the time.

Verstynen said the team鈥檚 solution for his dataset was simple, elegant and accurate.

鈥淸They] looked at the problem from a very different perspective than we had in the lab,鈥� he said.

Their solution could speed up the pace of brain mapping, something applicable in the real world: Mapping a typical brain and observing its connections can help further understanding disorders such as Alzheimer鈥檚 disease or schizophrenia, or help neurosurgeons determine the optimal trajectory for cutting into the brain.

Qualcomm Senior Director Rajesh Kumar noted the teams鈥� rigor and ability to achieve results in a short period of time.

鈥淭here鈥檚 a huge appetite鈥� for this type of event, Barth said.

Toneva said she would definitely participate again.

鈥淲e did pretty well,鈥� Toneva said, 鈥渂ut now we want to know how to do better.鈥�

More about the event 禄