BrainGrid+Workbench: Bringing High Performance Neural Simulation to the Desktop

Neural networks are often viewed as smart algorithms that can harness parallel computing to perform intelligent tasks. Paradoxically, real ” biological ” neural networks are massively parallel but so complex that it is difficult to map their requirements onto parallel computers. The BrainGrid project is building an open-source framework that will allow neuroscience researchers to move neural simulations from single-processor/single-core implementations to parallel, GPU (graphics processing unit) implementations. We have a beta version of our framework at git.io/braingrid, which we have used in our own research, simulating the development of 10,000 neurons over the course of several weeks. We have acquired the latest GPU hardware. We are now beginning to recruit external users. We then plan to work with those users to develop better documentation, wring out the bugs, and plan the next version.

Student qualifications

Comfort programming in C++ or Java. Desire to learn lots of new things fast. Interest in moving beyond “big-O”, to think about the match between algorithm and machine architecture. Brain!!

Student Responsibilities

• Learn some basic neuroscience.
• Learn some GPU programming basics.
• Learn about our framework and parallel algorithms.
• Attend weekly lab group meetings; help burn down the backlog.
• Test framework with new, simple, simulations. Improve our documentation.
• Engage in conversations about the next phase of BrainGrid development.

Resources and additional content

For additional information, visit the Biocomputing Laboratory website at http://depts.washington.edu/biocomp or the BrainGrid Github page at http://uwb-biocomputing.github.io/BrainGrid.