Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

[LINK] IBM simulate a "brain" with 500 billion neurons and 100 trillion synapses

5 Post author: drnickbone 21 November 2012 10:23PM

Recent article in The New Yorker:

http://www.newyorker.com/online/blogs/newsdesk/2012/11/ibm-brain-simulation-compass.html

Here is the research report from IBM, with the simple title "10^14":

http://www.modha.org/blog/SC12/RJ10502.pdf

It's nothing like a real brain simulation, of course, but illustrates that hardware to do this is getting very close. 

There is likely to be quite a long overhang between the hardware and the software...

 

Comments (22)

Comment author: Stabilizer 22 November 2012 01:55:29AM 3 points [-]

I think important caveats need to be kept in mind. From the New Yorker article:

I.B.M.’s Compass has more neurons than any system previously built, but it still doesn’t do anything with all those neurons. The short report published on the new system is full of vital statistics—how many neurons, how fast they run—but there’s not a single experiment to test the system’s cognitive capacities. It’s sort of like having the biggest set of Lego blocks in town without a clue of what to make out of them. The real art is not in buying the Legos but in knowing how to put them together. Until we have a deeper understanding of the brain, giant arrays of idealized neurons will tell us less than we might have hoped.

Comment author: gwern 22 November 2012 03:14:04AM 4 points [-]
Comment author: John_Maxwell_IV 22 November 2012 04:46:13AM 1 point [-]

The paper seems to indicate that the sim is running 388 times slower than real-time. I guess we're not 100% there hardware-wise yet? Good.

Comment author: RomeoStevens 22 November 2012 05:02:18AM 4 points [-]

this is worryingly fast IMO.

Comment author: drnickbone 22 November 2012 07:40:58AM *  1 point [-]

Thanks for this. The latest research report 10^14 already appears to be a significant update on that paper.

IBM now report roughly eight times as many simulated neurons and synapses, while the slow-down has gone from ~400x real-time to ~1500x real time. That works out at a factor > 2 in hardware improvement within a matter of months. They are using a custom hardware architecture and presumably there are still a lot of optimisations to be made. It can't be very long before this can run in real time.

As said in other comments, nobody knows how to program this yet...

Comment author: mapnoterritory 22 November 2012 09:19:01AM 1 point [-]

I actually never heard about non-von Neumann architectures. Anybody has some tip on a good source on this? Especially how this relates to biological brain architectures? Thank you!

Comment author: ChristianKl 29 November 2012 02:31:46PM 0 points [-]

It's nothing like a real brain simulation, of course, but illustrates that hardware to do this is getting very close.

They simulate model neurons. Those neurons they are less complex than the real neurons that we have in our head. The way in which real neurons change the amount of ion channels on their membrane for long-term plasticity is neither fully understood nor easy to simulate.