SoylentNews is people
SoylentNews
A new version of the NEST algorithm could dramatically reduce the amount of memory required to run a whole human brain simulation, while increasing simulation speed on current supercomputers:
During the simulation, a neuron's action potentials (short electric pulses) first need to be sent to all 100,000 or so small computers, called nodes, each equipped with a number of processors doing the actual calculations. Each node then checks which of all these pulses are relevant for the virtual neurons that exist on this node.
That process requires one bit of information per processor for every neuron in the whole network. For a network of one billion neurons, a large part of the memory in each node is consumed by this single bit of information per neuron. Of course, the amount of computer memory required per processor for these extra bits per neuron increases with the size of the neuronal network. To go beyond the 1 percent and simulate the entire human brain would require the memory available to each processor to be 100 times larger than in today's supercomputers.
In future exascale computers, such as the post-K computer planned in Kobe and JUWELS at Jülich in Germany, the number of processors per compute node will increase, but the memory per processor and the number of compute nodes will stay the same.
Achieving whole-brain simulation on future exascale supercomputers. That's where the next-generation NEST algorithm comes in. At the beginning of the simulation, the new NEST algorithm will allow the nodes to exchange information about what data on neuronal activity needs to [be] sent and to where. Once this knowledge is available, the exchange of data between nodes can be organized such that a given node only receives the information it actually requires. That will eliminate the need for the additional bit for each neuron in the network.
With memory consumption under control, simulation speed will then become the main focus. For example, a large simulation of 0.52 billion neurons connected by 5.8 trillion synapses running on the supercomputer JUQUEEN in Jülich previously required 28.5 minutes to compute one second of biological time. With the improved algorithm, the time will be reduced to just 5.2 minutes, the researchers calculate.
Also at the Human Brain Project.
Extremely Scalable Spiking Neuronal Network Simulation Code: From Laptops to Exascale Computers (open, DOI: 10.3389/fninf.2018.00002) (DX)
Previously: Largest neuronal network simulation achieved using K computer (2013)
(Score: 4, Interesting) by fyngyrz on Sunday March 25 2018, @12:19AM (8 children)
In order to create an AI with the objective of human levels of reasoning and consciousness, it should not be necessary simulate every neuron in the brain. A great deal of brain capacity is devoted to regulation of non-cognitive functions, such as heart rate, maintaining blood oxygen levels, balance, etc.
Further, some tasks that the brain does, such as balance, speech recognition and production and system regulation analogous to heart rate (hydraulic fluid levels, body temp, etc.) can be done very efficiently by very simple (comparatively speaking) hardware we already have.
So when the phrase "simulate the entire brain" is bandied about, we need to be aware that this is, in relation to AI at least, a considerable exaggeration.
Also – it may be that AI may be achievable with other methods than simply trying to duplicate the way we produce intelligence. If that's the case, the whole issue of "how many neurons can we simulate" might be right out the window. One hint in this direction is that birds have considerably different neural structures than other animals. [mentalfloss.com] The TL;DR is that bird brains achieve similar results to what our brains do in some areas, but they don't do it in the same way. If there are two ways, there might be three ways, etc. and this may well apply to other parts of the brain, and we're still talking about neurons as the basic building blocks. What if other building blocks can serve as well, or better?
(Score: 2) by takyon on Sunday March 25 2018, @12:32AM (4 children)
The stated objective in the short term may be to simulate medical issues, where the full system may be more illuminating. From the 2013 Riken article:
Creating a strong AI using this method could be a pain in the ass since the simulation speed would be very slow and you would want a way to train it. Yet it would be seemingly caught in a time warp. Assuming that the approach works at all for strong AI. Using a neuromorphic architecture to create something brain-like but slightly different in the way it operates could be a more feasible approach.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by tftp on Sunday March 25 2018, @02:09AM
(Score: 2) by fyngyrz on Sunday March 25 2018, @02:32AM (2 children)
That's why I titled my post as I did. My interest is tangental.
(Score: 0) by Anonymous Coward on Sunday March 25 2018, @03:08PM (1 child)
I was really confused. I have a programmable thermostat that takes two "AA" batteries that I change whenever the clocks need to go forward or back.
It has a tiny chip in it that lets me set the date and time. The NEST thing google bought didn't really strike me as a Neuromancer styled AI in some cloud of googles.
That whole robot home spy on you having sex so it can make the temperatures appropriate based on the VagiVibe app you synced to the smart home system never seemed to need a humain brain modeled AI to operate; it seemed to violate privacy without any effort or thought at all! And I thought humans did watch that stuff in the cloud?
(Score: 2) by takyon on Sunday March 25 2018, @09:18PM
https://en.wikipedia.org/wiki/Nest_Learning_Thermostat [wikipedia.org]
Already, you could see how that leads to some risks to privacy that could be exploited by criminals or LEOs.
It's also a hub for other devices from Nest or compatible devices. Such as security cameras.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by Ethanol-fueled on Sunday March 25 2018, @12:36AM (2 children)
Give that AI the sweet, sweaty taste of an American 7-11 Big-bite Hot Dog. It will twiddle its bits, stimulate its brain. Flood it with nacho cheese and it will flood your neural network simulation with good data as if your simulation were breathing pure crack cocaine smoke through a Vietnamese tiger cage.
(Score: 0) by Anonymous Coward on Sunday March 25 2018, @12:59AM
PS: your building in San Diego is pretty fucking lame.
(Score: 0) by Anonymous Coward on Sunday March 25 2018, @05:54AM
Ah ha...you and C-O are entwined! I knew it!