In my recent post, I commented on IBM's announcement at the Supercomputing Conference (SC09) in Portland, Ore., that they had simulated a brain with the number of neurons and synapses present in a cat's brain. It looks like the controversial statement of IBM being finally able to "simulate a cat's brain" (or however their original statements has been distorted) has been stirring some more comments. Henry Markram, the leader of the Blue Brain project at EPFL, Lausanne, sent an open letter to IBM CTO Bernard Meyerson, along with several media (UK Daily Mail, Die Zeit, Wired, Discover, Forbes). One big question is: was Modha's statement somehow distorted? Did he actually simply claim that IBM simulated a system that has the same number of neurons of a cat, as opposed to simulate "the cat's brain?". This is an important distinction. Anyway, Neurdons must know, so here it is! Enjoy!
Cat fight over blue brain
| November 24, 2009
About Massimiliano Versace
Massimiliano Versace is the director of the Boston University Neuromorphics Lab. The lab focuses on the study of biological intelligence with the goal of embedding the derived fundamental principles in bio-inspired computers and robots. His research interests are focused on neural networks – in particular applied to spiking-based neural models of learning and memory in the cerebral cortex. With a few colleagues, he founded Neurala LLC in 2006 to commercialize brain-based software. For more info, visit his website
Comments: 4 Comments
I knew I was a Mac user for a reason! Oh, IBM, will you ever learn….
I too agree with Markram on his critique of Modha. I think to be able to accelerate neural network simulations is necessary but not sufficient for SyNAPTIC success. The main distinguishing factor from IBM’s effort and Markram’s blue brain project will be the building and hardware-incorporation of an [evolving] ‘robust’ biologically-inspired model of a rat/cat/human brain that can demonstrate “visual perception, decision and planning, and navigation”, and other ‘appreciable’ behavioral competencies.
I just published an email I got today from Dr. Markram over at http://www.Technology-Report and I’m really interested in your reaction to his critique of IBM’s amazing claim.
Hi Joseph,
great idea in inviting Markram to your blog. Henry is actually a very welcome and frequent visitor at the center where Neurdon was born, CELEST. He points out a number of true statements (among which Eugene’s Izhikevich simulation in 2005 that, from the neuroscientific perspective, wes of a larger scale and more interesting that the current one).
I would like to take a step back from the controversy, though. I am not sure how much Modha’s statement has been, either on purpose or voluntary, distorted to fit various interests (IBM, reporters, etc). I have read the paper, and I was surprise to notice that the architecture chosen by Modha and colleagues to implement in C2 (described in Figure 1) is impressively similar to the one I published in 2008 in Brain Research, and started presenting in various conference in 2006. In fact, I recall talking to Modha about it in SfN in DC, I believe, in 2006!
I am therefore very familiar with the model and its limitations. I simulated a smaller scale of Modha’s architecture with Steve Grossberg to study how stable learning of synaptic connections can develop in the presence of STDP. I would love to see, if this is the direction that Modha is undertaking, their take on this essential and very complex issue.
Finally, I would like to point out that, despite I have to agree with Henry on the neuroscientific impact of the IBM work (not very big, I have to acknowledge), I think that the real value of the contribution is purely technical, namely…skip the first few pages and go read the details of their simulations. They successfully handled 147,000 processors, 144 terabytes of memory, by getting only a .3 percent deviation in workloads across cores. That, I believe, is the contribution of the work…. it sounds less sexy than simulating a cat brain!