By matching video of mice’s movements to their brain activity, Carnegie Mellon University looks to better understand and treat neurological diseases.
“Before we got involved with Oracle, things were difficult. It took a huge load off of our back to be able to have this additional resource. It's user friendly, it's whatever we needed it to be, and it's flexible, enabling whatever we want to do experimentally.”
What separates man and mouse? Less than you would think. “Natural behavior” is a profound concept that promises to transform our understanding of the animal and human experience alike. The term is defined as “behavior that animals exhibit under natural conditions, because these behaviors tend to be pleasurable and promote biological functioning.” Accurately establishing what is a positive and negative experience for an animal in these terms will be key to improving animal welfare, and could also provide insights vital to preventing and treating conditions such as Parkinson’s Disease and obsessive-compulsive disorder, and gaining a deeper understanding of humanity’s evolutionary roots.
To study this natural behavior, the CMU team recorded mice with four cameras 24/7 for several weeks, while using 386-channel electrodes to monitor the mice’s brain activity. And that’s where the computing challenge came in.
Due to the complexity of natural behavior, Carnegie Mellon University Assistant Professor Eric Yttri believes that researchers have been using inadequate methods to observe the behavior of mice, and that these approaches fail to capture what the brain is actually designed to do. New machine learning research by Yttri and CMU Graduate Student Alexander Hsu open the door to measuring natural behavior in a more meaningful way.
To apply this new approach, the team at Yttri Lab needed easy and fast access to powerful processors and considerable storage in a computing environment customized to their exact needs. While CMU offers an on-premises supercomputing service, gaining access is time-consuming and users can’t control the software packages installed. Moreover, the failure of one compute job can cause delays of several days with this on-premises solution, so the team was looking for a faster, more flexible alternative that was also cost effective.
Because we're continuously recording, we have to render down the data captured that day, before the next day starts. That requires a huge computational load that Oracle provides. It helps tremendously in terms of the speed of processing.
Why CMU chose Oracle
With the help of Oracle for Research, CMU used Oracle Cloud Infrastructure (OCI), running multiple Oracle VM.GPU3.1s plus VM.GPU3.2, in addition to 20 TB block-volume cloud storage and 10 TB block-volume cloud storage. The team streamed 3 TB of data onto OCI per day during recording sessions.
The researchers found OCI GPUs and storage significantly more accessible and customizable than CMU’s on-premises supercomputing service. Compute jobs also ran fast, since the team wasn’t sharing these resources with their colleagues on campus. The flexibility of OCI virtual machines also enabled them to customize a computing environment to their specification.
Oracle also provided tech support to help ensure the project progressed smoothly. The Oracle for Research initiative helps researchers by providing cloud credits, hands-on technology consultations, introductions to peers, and more.
CMU used Oracle GPUs to process both data streams—the video of mouse behavior and the electrodes monitoring brain activity. Pairing these recordings of neurons and actions let researchers build a comprehensive, neuro-behavioral database. The raw behavioral signal contained a lot of data that was not useful, so Oracle technologists helped the team to build automation to remove the vast amount of data that was unneeded.
This new database effectively bridges the activity of individual neurons and neural networks with each individual action demonstrated by the mice, casting new light on how the brain translates thought into action. The CMU team found that brain functions are more interwoven than previously thought, including individual actions not typically associated with activity in one specific area of the brain.
Such neurobehavioral datasets have been needed in this domain for decades and should be useful for other cognitive scientists looking to study neuropsychiatric conditions, and even for engineers who want to reverse-engineer artificial intelligence (AI). Analysis of this dataset will form the cornerstone of the Yttri Lab’s future research, and likely that of dozens of other groups.
About the customer
Carnegie Mellon University is a private research university in Pittsburgh, Pennsylvania. The university is the result of a merger of the Carnegie Institute of Technology and the Mellon Institute of Industrial Research. The university has seven colleges and independent schools and is known for many firsts in computer science, such as creating the first computer science, machine learning, and robotics departments.