"It seems biology has found a way to distribute information so that if you lose small components, you don't lose the memory" An acronym must be produced for this, it is IT law. I suggest; R.A.I.N (redundant array of independent neurons) ©2015 My guess is that the brain uses a RAIN 6 configuration.

Wow! This is actually a ComputerPile

in 2065 we'll laugh at how these computers were big as rooms in 2015 just like we now in 2015 laugh about how computer were big as rooms in 1965

now, add 99 million more

There was an experiment a year or so ago (maybe two) which connected the brains of two mice. They even did it over a network, such that impulses from one brain were picked up by a sensor and conveyed to the brain of a different mouse. I want to see a similar approach taken, except connect a mouse brain to one of the SpiNNaker boards. Start with a 'blank' network in the board, and connect the mouses brain to it - preferrably from many points on the mouse brain, and starting with the youngest mouse possible. I would also like to see more videos about this project, concentrating more on the software than the hardware aspect. How do they encode the neural network? For the event-driven architecture on each core, did they develop a low-level system or are they using maybe a stripped down existing kernel (maybe one of the academic kernels with proven correctness?)?

That sounds like an amazing project. It is hard to believe that they will be able to simulate the number of nerve cells inside a mouse brain. Though, the question I have is, how fast will the simulation be, compared to the real life processing speed of an actual mouse brain?

I have been doing this and thinking about these things for 7 years now, and his machine make me so envious. I think that event driven is the way to go when simulating a "brain" or consciousness, looking forward to hearing more about this in the future.

Wow this is amazing, I wish you the very best with this project Prof. Steve Furber.

Has the PHD student not heard of cable management?

PLEASE more on this!

i love what you guys are doing. it is very difficult, but also very useful. cheers.

I'm sad to not see professor Steve Furber around this channel anymore :'( please bring him back, he is one of my favourites!!!

Million cores = 1% of human brain? All things considered, two more orders of magnitude (or like 7 if we go in factors of 2) isn't _that_ far off anymore. The only problem is the overall size of the machine in question. But given that a large portion of our brain is made for navigating the world, perhaps that much isn't even necessary to mimic cognitive capabilities of humans.

On Sunday, ***** posted about brains made of skin, now this about brains made of ARMs! The internet is a scary freak show.

its 1,000,000 you forgot a comma

Thank you for these videos. I'd have no access to all this otherwise! Amazing stuff

Finally a very interesting video here. Thank you

That was a super interesting video thanks!

This is really cool!

To Coop772: I'm an european engineer working at an American chemical company. Yes, I'm used to indicate pipe diameters in inches because of the ANSI standard we use. Because of this I really think in inches when talking about pipe sizing (everybody in our company does). However, wall thickness is back in mm again and lengths in meters.

0:23 I take personal offense to that. My brain is the equivalent of AT LEAST an even dozen mouse brains!

+Computerphile, would it be possible to make a video about what this computer can do and how it does it? I feel you only hinted at what it is capable of and this is just so fantastically incredible (for now) that it deserves more attention. The fact that we are up to 10 mouse brains is for me quit an achievement.

Memory doesnt get deleted like that because hundreds of neurons dont store a memory but its the symphony of millions of neurons firing. in a rhythm We are so far away - its like we are playing w mud still trying to make transistor. We need to be studying the nonlinear equations that govern the neuron and their dependence on the array of receptors at the connections. Once we know more about that then we can apply it to the latest array of processors.

Imagine the amount of power needed for all the processors to potentially simulate 1% of the human brain! Our brains run at 100% and only need an apple and a coffee in the morning to function xD

Holy moly, this is gorgeous. This makes me soo giddy!

Anyway, imagine going back in time and showing the plans to someone in the 1960s. "These ten racks, in the future, won't just house one central processor, but _a million."_

Coop772: (sorry, for some reason I can't reply directly to your question) Here in the UK, we have a particularly strange mix of units. Feet/inches/pounds etc. are called imperial measurements, I think because of long disappeared British empire, but we increasingly use metric. I know for sure the metric system comes from France. It's now even become illegal to only display imperial values at shops as we try to standardise with the rest of the European union - a bit over the top but lets face it if you put what you know and love aside it does make more sense when everything is more and more computerised. Units that go up to 12 or 16 can make perfect sense when you do mental maths however. Now, schools teach pretty much only metric, apart from miles - that's still a standard measurement. Speed limits and distances are all in miles. Every car I've had, including German ones and Toyotas built both here and in France, all are fitted standard with a mph speedo, and measure fuel consumption in mpg. I guess if we started measuring in kilometres per litre, with our high petrol prices, we would all die of dehydration from non-stop crying. I would however prefer to be able to speed along at 100km/hr rather than the pedestrian 62mph, even though they are the same! So you end up with older people still using pounds and inches, but more and more, especially for units that you directly experience yourself - the weight of your luggage, the length of your toothbrush, etc. metric is taking over. But of course there are always exceptions, the one I can think of is floorspace because it sounds bigger in square feet. My house is leet square feet (really - the builders plans say 1337 sq ft), but in meters it would only be 130. Hope this helps if you ever read it!

Amazing project! I wonder why they didn't use FPGA's instead. Looks like positive impact on both design process and final processing capabilities to me.

He's such an eloquent speaker :-)

Outstanding! I'd love to be able to contribute to the project.

Those circuit boards look awesome.

I wish this video was 18 hours. Detailed content about this project, it's so fucking interesting.

A spinnaker is also a sail hence the image shown when he starts his description of the project.

I want to know more about this!!!

I had a brief but good chat with him at cafescientifique on the subjuct id love to get in volved in this project more

0:07 Damn spinnakers!

Wait, did he just say the human brain is roughly 1000 mouse brains? I guess I'm okay with that.

To the coop772 guy: Yes we use both meters and feet and inches. We use the unit that is most similar. Like how your pinky is like a centimeter wide, but your knuckles are an inch long. Much better than saying a pinky is 1/3rd of an inch or that your knuckles are 3 centimeters

I'm actually floored that it only takes a million to get to 1%. Only 100 million to emulate (theoretically) a human brain! Amazing.

just cool, when is it going to be completed?

Wow. What a cable management.

4:55 aaaah that cable management....

Obviously they must have started this project quite some time ago, but what i'd be interested in is seeing a large cluster computer built around the IBM SyNAPSE chip, which has one million neurons and 256 million synapses implemented in hardware on a single chip.

Fascinating. When comparing the human brain to a computer you can see some similarities. For starters we look at memory. Computers have long term, short term, and permanent memory. Humans brains have similar characteristics.

That are allot of chips. But good to see that ARM is so useful. Most consuments only get to meet them in their printer or their tablet.

this is amazing I hope this technology advances for society's advantage

1% of the full human brain capacity! It's huge! As somebody would say "we are only 7 doublings away from the whole thing". It seems like genome sequencing 7 years before first successful sequencing

AH MANCHESTER. THE WORLDWIDE CENTER OF ARTIFICIAL BRAIN BUILDING

This is so damn cool...

An update on this project is mandatory now that the million mark has been acheived !

In 40 years we'll be commenting on how ridiculously large of a machine we needed to simulate a brain, same as we are surprised at the size of currently 40 yr old calculator. Technology is great.

1965 they weren't measuring transistors in nanometers, they can't scale down infinitely, hence the need for multiple cores, hyperthreading and bus speeds. My prediction would be that they'll need supercomputers to efficiently hard code these multicored super comps and we'll go back to algorithms making things faster rather than processing. The brain works because of its ability to rearrange itself on the fly

What are the properties of each of the million neurons and what are the connections between them? An Aplysia feeding network can be described by a dozen neurons but they are all different and inhibit or excite each other in specific ways.

Those damn Spinnakers with they green backs and grey processors!

If you would just simulate a simple network, and add some I/O, could the network "learn" a few things, if you would run it long enough? There are some programms, which can learn some games like platformers or racing games, but i think it would be much harder for a virtual network to learn how to decode something like a video signal.

"If you were plowing a field, which would you rather use? Two strong oxen or 1024 chickens?" -Seymour Cray

Amazing stuff, but why link them in a toroidal surface? also I am not quite sure I understand what that means... Is it to minimize the distance between any two cores?

I wanted to click that "How Computer Memory Works" box but there's no annotation fyi. Just ending up searching. Cool vid!

Really dumb question: A lot of simulation work is done on high end GPUs as far as I know, because GPUs essentially have sometimes thousands of cores which can be used for parallel workloads. I assume it's because a CPU is more versetile in it's instruction set or because GPU cores are much lower clocked ore something like that. But I also assume that there is some other much more glaring issue?

All your cores are belong to us. :P great video.

Can someone explain the toroidal hexagon signaling arrangement? I’m having trouble understanding what it would look like. (I have some experience with hardware signaling, timing, and addressing so it’s not quite necessary to ELI5)

They do have some knowledge about human memory, long term memory takes at least 2 years to fully form. The long term memory is located over the entire neocortex. This is basically the entire outside of the large human brain.

damn spinnakers

Coop772 I can't reply to your comment for some reason. So I'm doing it this way. The person in the video is clearly English and of an age where he would have grown up with the Imperial system. As he got older and the Metric system took over, there would be the tendency to swap between the two systems depending on the thought process at the time. As a Brit myself and of the age of 52 (in November lol) I too do the same thing, Although the majority of my schooling was in the Metric system (From 1972 onwards) I can visualise dimensions easier in the Imperial but often use the Metric system for many other things. We haven't been using the metric system long enough for this kind of thing to disappear.

Why ARM? Why not clusters of fpga's?. I can see some Spartan 6s but I assume they are just do interconnect logic/routing. I am glad to see that something like this is finally built but the way I imagined is instead of using processors we could use fpga clusters with an sram under each fpga (on the backside of the pcb). You could model orders of magnitude more neurons that way.

10e9 simulated real-time neurons is impressive but what is the number of connections between them? It would appear this is the limiting factor to the complexity also.

I'm just being cheeky but we should be parading Steve Furber around like Woz, telling his story over and over worshiped and transported from glorification to glorification. i suppose he Steve still works, good on him ! a real british hero for the ages. every day of his work is gift from god.

How did they connect so many processors onto one circuit board? And does anyone know where I can learn more about comptuer hardware like that (best books? etc). I'd like to tinker and make my own 1000 core (not a million!) circuit board, it sounds like a lot of fun.

So... Its been... One year? Two years? Since this video went up. I'm curious as to what progress has happened in the year since then.

Question: What would be, regarding the laws of physics/the universe, the most powerful computer ever (in terms of how fast it can compute)? Does the Bekenstein bound and the Bremermann's limit describe said computational limit?

Very cool research! Maybe the ip address of the robot mouse should be pixelated out?

Coop772 In most countries where the metric system has been in use for generations, people use it exclusively. UK however only adopted it recently and reluctantly so it's normal for them to use either imperial or metric units depending on the situation. I'm guessing he just got so used to both that he converts between units freely without even thinking about it. There probably exist 2 sets of specifications for the equipment, one for Americans and one for the rest of the world lol

These guys are going to map the human brain, and hopefully bring us Nerve Gear!

did he say we lose a neuron every second? for real? I'm scared af now man

Do you know what the floating point performance of the machine will be?

19 inches = 48.26 centimeters Subtitles with metric translation would be awesome to get immediate feel for measurements. ;)

Impressive.

holy damn i get all tingly :O

So they would need an entire mini warehouse to model a human brain? Amazing.

i have a question. are these running like a 1 million core cpu, or is it wired differently?

YES!!

How about an update. Is the project finished, how is it going? Did they get to a million cores? Brady?

just wondering how much ram and flops does the thing pack?

The question is, does Arkham Knight still stutter on 100,000 ARM processors?

any idea what language they are using for the software ?

Wait wait wait! They're actually using ARM cores that are commercially available and making networked boards themselves. If the Manufacturers themselves start to make networked boards I can easily foresee a 100-fold reduction in size and maybe 10-fold reduction in power consumption on their first iteration!

Current neuroscience (FMRI) is becoming aware that the human brain process is not just the neurons alone. What then does this computer model hope to achieve.

that little robot got a big brain

It seems to me the human brain does not store memories comparable to the way a computer does it, nor in a redundant way. If you assume redundancy, you are still assuming the same storage model. It looks like memories are reproduced from a more fundamental source every time they are needed. This makes sense when you consider the quality of memories DOES deteriorate over time, as the brain gets older and damaged.

I wonder. As long as this brain is not required to work with human real-time stimuli, what's the problem with it being 1% of a human brain in terms of volume? Wouldn't it be possible to go to ~100% by running it multiple times with different state? A rather common trade-off between computation space and time?

Interesting project, but how will this model quantum effects in the brain? This still looks like a big fancy calculator, no matter how complex the program. You can't model quantum effects using pseudo-random number generators or any other deterministic processes.

Stwve Furber = Legend,

Sir, are you building the #Skynet alpha there?

18 cores x 48 chips x 24 boards x 5 frames x 10 racks = Merry Christmas!

The missing comma in the title bothers me

imperial or metric, I can't work betweensies!

That number in the title. 1000,000

Crazy idea I want to see it control a robot mouse interacting with some real mice.

What OS are you running on those?

I wonder if such arrays could be sunk into mineral oil for cooling rather than using fans.