CMU team aims to unlock brain’s algorithms

Raghunandan Avula Feb 14, 2016

As humans, we have invested in the development of computers to automate the world around us in order to increase efficiency and the reach of human understanding. While computers can seamlessly make a large number of computations very quickly, the machines are little competition for the the biological machines we find in nature, like the brain. A developing research area in computer science is focused on better understanding biology and using the insights for algorithm development.

Following this path, Carnegie Mellon researcher Tai-Sing Lee, a professor in the Computer Science Department and the Center for the Neural Basis of Cognition (CNBC), will attempt to reverse-engineer the brain in order to try and reveal its learning methods and apply them to advancing machine learning algorithms. The project will span five years and will cost roughly $12 million.

The human brain is made up of many small units called neurons that send messages to each other through a massive network of electrical impulses. These messages mediate all the functions the brain conveys to the rest of the body, which range from actions such as interpreting the scenes in a video to moving your hand. While scientists don’t fully understand all the intricacies in a neurons’ functioning, biologists have gained a breadth of understanding in recent decades.

Recent data has been starting to uncover that these neurons are highly interconnected and may even participate in feedback loops, leading to a very complex system. While biologists have been hard at work uncovering the secrets of the brain, computer scientists have been busy with the advancement of machine learning, which has seen a remarkable growth within the past decade.

The main goal behind machine learning is to use a computer to interpret patterns with the data it receives to generate a computational model and use the model to make further predictions given new data. The models can also evolve with new data and become more robust. Machine learning algorithms have been responsible for many recent advancements such as genomic analysis, the power behind large-scale big data, and even self-driving cars.

Recent development of new technologies has made it possible to collect large amounts of data regarding the brain’s activity, and more importantly, to be able to analyze that data and make important conclusions. One of the main goals of Lee’s project is to create a large data repository which represents neural circuitry. In order to do this, Lee is collaborating with investigators at Cold Spring Harbor Laboratory, MIT, and Columbia University.

This repository will then be available for other researchers around to world to conduct their own analysis. In addition, Lee will collaborate with Sandra Kuhlman, assistant professor of biological sciences at Carnegie Mellon and the CNBC, and Alan Yuille, the Bloomberg Distinguished Professor of Cognitive Science and Computer Science at Johns Hopkins University, to analyze their data.

A technique already used in machine learning is one of neural nets: algorithms based on simple units like neurons that share information, or in this case data, with each other and make computations. To better understand biological neural nets and their feedback loops, Kuhlman will utilize 2-photon calcium imaging microscopy to record signaling of tens of thousands of individual neurons in mice as they process visual information.

The modeling of thousands of neurons at a time is a scientific achievement that wasn’t possible a few years ago as only a few neurons together would be monitored at once.

“AI has gone from failure to failure with bits of progress,” Yann LeCun, director of the Center for Data Science at New York University, said in a Carnegie Mellon press release. “This could be another leapfrog."

Using this massive dataset, the researchers hope to unveil intricacies in neural nets. Their goal is to then apply their new knowledge to machine learning algorithms that would enable them to develop a model given a fewer number of examples. This holds great significance for the future, since machine learning is a burgeoning research area within many lab communities.

The idea being that if we can somehow understand the brain, we can understand how to replicate it, and our machines will become capable of learning on their own. The dual goals of this research project are very in tune with Carnegie Mellon’s interdisciplinary culture.

This research is made possible through funding from the Intelligence Advanced Research Projects Activity (IARPA) through its Machine Intelligence from Cortical Networks (MICrONS) research plan.