Computational Analysis Methods and Issues in Human Cognitive Neuroscience
Google Tech Talk January 14, 2010 ABSTRACT Presented by Bradley Voytek. There is a massive, relatively uncoordinated effort underway to map out the relationship between brain and behavior. Human neuroimaging experiments abound with approximately 30000 neuroimaging studies performed in 2008 alone. Most of the data from these experiments are analyzed on an individual desktop or small, local cluster. Neuroimaging data contains information about neural activity in both time and space and can easily exceed 1GB per subject. In order to analyze the functional properties of neuronal networks these data can be decomposed in a variety of ways (behavioral condition, principal and independent components, phase and frequency components, graphs and digraphs, etc.). This exponentially increases analysis time and database sizes creating bottlenecks in the analysis work flow. I will discuss a variety of neuroimaging methods in terms of the sources of the signals measured, what these signals actually inform us about how the brain gives rise to cognition and behavior, and how this information can inform medical diagnosis and treatment. Furthermore I will highlight how advances in computational processing have improved data analysis and discuss the computational roadblocks that impede research progress.
May 28th, 2010 at 10:22 am
It was a great lecture nonetheless.
May 28th, 2010 at 10:27 am
he talks fast.
distributed computing is not applicable to brain sim.
samplingrate isn’t the problem, (cell)-resolution is.
CERN/LHC is collecting from many sensors at a time.
May 28th, 2010 at 12:04 pm
Oh, don’t get me wrong, there have been many advances. But that wasn’t exactly the point of my talk. I wanted to highlight the issues.
May 28th, 2010 at 12:17 pm
Very interesting talk. Thank you for it. Since you are having issues with the computation and sampling, distributed and/or parallel computing may help to gather more data at any particular time but since I don’t really know the specifics, I cant even attempt to suggest a path to a solution but it may lie in that direction.
May 28th, 2010 at 2:22 pm
The cpu cycles needed for simulating the human brain are crazy huge. The most powerful supercomputers in the world can barely simulate the brain of a cat. I guess something like Folding@Home would help, but yeah, cpu time is a major issue to overcome.
May 28th, 2010 at 2:57 pm
yes it did. i’m dead now.
May 28th, 2010 at 5:55 pm
True there was/is an attempt at the moment to do just that, I think it’s IBM. But you are right no supercomputer at the moment could help in simulating a human brain. I was more talking about the sampling of the human brain at any given time not simulating it, since this was one of the obstacles the speaker talks about in the presentation.
May 28th, 2010 at 6:15 pm
He might be distracted by the technical audio problems.
May 28th, 2010 at 6:28 pm
no it didn’t
May 28th, 2010 at 6:40 pm
Have we made “any” advancements? This guy keeps reiterating the idea that we know nothing about the brain and all the science is just a pure crapshoot.