Slashdot Mirror
fMRI Data Reveals How Many Parallel Processes Run In the Brain
New submitter xgeorgio writes: From MIT Technology Review: "The human brain carries out many tasks at the same time, but how many? Now fMRI data has revealed just how parallel gray matter is. ... Although the analysis is complex, the outcome is simple to state. Georgiou says independent component analysis reveals that about 50 independent processes are at work in human brains performing the complex visuo-motor tasks of indicating the presence of green and red boxes. However, the brain uses fewer processes when carrying out simple tasks, like visual recognition.
That's a fascinating result that has important implications for the way computer scientists should design chips intended to mimic human performance. It implies that parallelism in the brain does not occur on the level of individual neurons but on a much higher structural and functional level, and that there are about 50 of these. 'This means that, in theory, an artificial equivalent of a brain-like cognitive structure may not require a massively parallel architecture at the level of single neurons, but rather a properly designed set of limited processes that run in parallel on a much lower scale,' he concludes." Here's a link to the full paper: "Estimating the intrinsic dimension in fMRI space via dataset fractal analysis – Counting the `cpu cores' of the human brain."
That's a fascinating result that has important implications for the way computer scientists should design chips intended to mimic human performance. It implies that parallelism in the brain does not occur on the level of individual neurons but on a much higher structural and functional level, and that there are about 50 of these. 'This means that, in theory, an artificial equivalent of a brain-like cognitive structure may not require a massively parallel architecture at the level of single neurons, but rather a properly designed set of limited processes that run in parallel on a much lower scale,' he concludes." Here's a link to the full paper: "Estimating the intrinsic dimension in fMRI space via dataset fractal analysis – Counting the `cpu cores' of the human brain."
No the brain runs a very efficient version of systemd that has replaced ascii bash through a binary remote code execution system which is much more efficient and simple.
The brain is an analog computer. The notion of parallelism is fundamentally different for an analog computer... In a sense, every single neuron is operating independently and in parallel with the rest. Describing it in terms of parallel processing with digital CPUs makes no sense.
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion.
the screen displays either a red or green box on the left or right side. If the box is red, the subject must indicate this with their right index finger, and if the box is green, the subject indicates this with their left index finger.
I'm color-blind you ignorant clod. Green, brown, yellow, red, whatever ...
Typically, fMRI machines divide the brain into three-dimensional pixels called voxels, each about five cubic millimeters in size. The complete activity of the brain at any instant can be recorded using a three-dimensional grid of 60 x 60 x 30 voxels.
Is this a fine-enough resolution? If we used 1mmx1mm, would we see more than 50 "areas of activity" at one time? Or are we assuming this because that's what we have available right now?
"Transparent" is a shit show that trades on every stereotype going. A man in drag is NOT a transsexual.
Are well known to be massively parrallel and occur at the level of individual neurons.
It looks like this is just muddying the waters between functional units and internal parallelism
It has no relationship the common usage of the term in computing, a far better way of phrasing that would be tasks.
... the answer is one.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Oh Please Edge Detection and Motion Detection. Are well known to be massively parrallel and occur at the level of individual neurons. It looks like this is just muddying the waters between functional units and internal parallelism
A vision system has multiple tasks to perform, low and high level tasks. The edge detection and motion detection that you describe are primitive feature extraction operators, low level tasks. Cognition, the interpretation of these features that allows the building of a model of what is being seen is something very very different, a high level task.
For example determining the magnitude of an edge and the direction of an edge by looking at a pixel and its immediate neighbors is a very simple mathematical operation. And because of the locality of the inputs, pixel and immediate neighbors, it lends itself to massive parallelization. Interpreting a large number of edges over perhaps a large part of the field of view to recognize the immediate environment using a memory of stored models and templates has completely different computational requirements and an entirely different opportunity (or relative lack of it) regarding parallelization.
Opens list of processes
Finds "Repeat partial song on loop indefinitely"
Ends process.
1. Yes, the brain is massively parallel and "analog" - BUT not every neuron forms a distinct cognitive function (FBNs) and neuron do fire in pulses/spikes (almost binary) rather than continuous (analog) outputs.
2. No, the article does NOT identify 'cpu cores' in the brain. It uses this metaphor (stated clearly in the paper as such) to point out the level of parallelism needed to run anything remotely similar to the complete functional 'package' in the brain.
3. The resolution of modern fMRI is at 3mm^3 (30K-50K active voxels) but this has to do more with the localization of the activations and less with the inherent complexity (dimensionality) of the spanned data space. In other words, in this work e.g. the visual center is detected as activated or not, regardless of how fine the resolution is.
4. The fMRI captures the complete 3-D brain volume, hence the detected activations include all the "always on" circuitry like respiration, cardiac rhythm, etc. Cognitive processes are only a few of these activations and are identified by experts when looking at the actual activation maps.
5. The methodology is completely data-driven and it includes two very popular non-parametric approaches: one is ICA for blind-source separation (measuring how many components are needed to describe the data) and the other is dataset fractal analysis (estimating the intrinsic dimensionality of any dataset). In both cases, the maximum number for such a plain visuo-motor task seems to be around 50.
6. The number 50 is only indicative, as it is measured for specific fMRI visuo-motor experiments. In intense cognitive situations, e.g. a pilot trying to land a plane on an aircraft carrier at night with bad weather, this is probably much higher - but in he same order of magnitude. On the other hand, when very small activations are ruled out (pre-processing by voxel smoothing), this number becomes much lower.
7. Currently, we have no idea how to develop a fully functional "brain" just by putting together 10 or 50 or even 1000 parallel processes. The simple idea of the data-driven approach is to point out that we should focus on independent -neural networks- rather than -single neurons- when trying to simulate an actual brain.
8. The current state-of-the-art neuromorphic chip by IBM provides just about 1/3 of a single voxel with 1/40 of neuron synapses within, so it is imparative to see how we can use these resources the best we can.
I hope these hints make things a bit clearer now :-)
"Abashed the Devil stood, and felt how awful goodness is..."
I think they mean the human brain, not brainfuck.
Then you might be interested in reading this, which describes how it might all work, and how an (actual) AI could be made to work.
I've fallen off your lawn, and I can't get up.
...to the one using the hammer, there is a tendency for everything to look like a nail. Identifying fMRI correlates may not actually indicate the number of cognitive components in play, any more than counting the number and location of gasoline stations tells us much detail of what people in a city are doing. At most, it gives us some useful hints.
If you want to emulate a brain with chips, you have a major obstacle to overcome; the fact that chips don't change dramatically over time as they acquire experience with the world through a coordinated set of sensory-motor systems. You would not just need the 50 or so high-level processors that are dedicated to specific tasks, linked together very specifically, you would also need the entire system to be able to rewire itself at both microscopic and macroscopic levels based on experience. Without living organisms intrinsic ability to remake the system in real time at multiple levels of structural organization, chips will always just be chips trying to imitate the brain. They will need to be able to learn and grow, just like brains, or they will always be cheap imitations.
A brain is a terrible thing to waste... Mind? That's debatable.
While driving about a dozen years ago, I thought: "Wow, I'm thinking."
Then I thought: "Wow, I'm thinking the thought: "Wow, I'm thinking.""
Then I thought: "Wow, I'm thinking the thought: "Wow, I'm thinking the thought: "Wow, I'm thinking."""
Fortunately I didn't crash.
However, I couldn't get to the next level without my mind drifting elsewhere.
bullshit.
the limitation of resolution of current fMRI tech shows ~50 funcitonal macroprocesses, improve the resolution and that number will change. each of those macroprocesses is microthreaded out with extensisive superscalar processing and highly effective branch prediction. in addition much processing is distributed out to leaf nodes outside the brain proper.
identifying red and green boxes? Bah, try sorting apples (size, finish, remove debris, leaves, etc) loading each category into moving crates/bins, chewing gum, thinking about if the kids did their homework, talking to the person next to you about their love life, wondering why you didn't get flowers too, remember to do pelvic floors and how much longer until a bathroom break, gee my feet hurt, maybe I should get new shoes, gee those shoes I saw last week were nice but payday, omg bills etc
* If the new process paused because it was
* swapped out, set the stack level to the last call
* to savu(u_ssav). This means that the return
* which is executed immediately after the call to aretu
* actually returns from the last routine which did
* the savu.
*
* You are not expected to understand this.
*/
(credit to http://cm.bell-labs.com/cm/cs/...) And yes, let me forestall a lot of comment -- as the link above mentions, the code associated with this comment in the v6 UNIX kernel was wrong.
So, I'm not gonna be able to simulate the brain on an ATTiny85, then, am I? Not even at 20 MHz?
The notion of parallelism is fundamentally different for an analog computer...Describing it in terms of parallel processing with digital CPUs makes no sense.
came here to say this...
that misunderstanding is an inherent problem in computing and "ai" i fear
our brains are not "like computers" in how they work
Thank you Dave Raggett
I feel like ascii-bashing systemd and some brains now.
Coarse graining applied to brains. What else is new? Recognizing separate functional blocks in an undamaged adult brain is one thing. Creating an equally plastic system from reconfigurable logic exhibiting dynamic parallelism is another. A computer core which survives, reroutes and scales accordingly after a direct micro-meteor impact would be nice.
That's all interesting but how many things can a brain do at the same time CONSCIOUSLY? Many studies point to the same number: 1 (one).
-- Cheers!
The scanner measures the hemodynamic response function. The brain only sends oxygen-rich blood to the regions of the brain that need it. I guess this restricts power consumption, heat sinking requirements, and so on, a bit like the power limiting circuits on a processor. It is likely that the power regulation is a lot less fine grained in the brain than the thinking process itself. So if you had two separate regions that were fed by a single blood supply, you would not be able to distinguish them. In practice, I expect the processing regions and the blood supply regions have fuzzy borders, and there is a limiting return in micro-managing the blood supply.
I do not understand the gas station analogy. To me, this is more like trying to tell how many people are in a building by how many lights are on. You can subtract the permanently on lights from stairwells and corridors. You can then assume a light that goes on, and goes off may be a single person, or a meeting, or a group of people who all come and go at different times. Subtract the lights that are permanently on, and you would expect the person could to be at least the remaining light count, because several people may use the same light.
Hey, it's a start.
Ewww... No thanks.
Mod parent up AND consider:
a) remember that the use of Independent Component Analysis (ICA) is appropriate for linear processes and therefore must necessarily be, to an unknown degree (until you actually know the underlying distribution), an approximation ie. the more unlinear the process, the less ICA accurately reflects the underlying processes; and
b) the actual processing methodology of the brain is unknown, heck, we do not even understand the encoding used by the brain.
So the article really rests on the assumption that the brain is composed of linear processes operating like a modern digital computer.
Ummm ... no.
"Consensus" in science is _always_ a political construct.
Brodmann already counted the CPUs of the brain. They are called Brodmann areas. BA17, for example, is primary visual cortex. BA45 is Broca's area (speech). There are about 50. They are defined by differences in the micro-cellular architecture of the area. Most areas of cortex look roughly the same, but there are many differences, for example the input layers of primary sensory areas are larger than in other areas. Some areas have large output layers, or more inhibitory cells, etc.
The brain does have many distinct areas, asynchronously operating, highly connected with both local and long distance connections, and the areas themselves are composed of a rich mosaic of different cell types that continuously self-regulate, process information, and adapt.
How does this apply to phrenology?
Tits. Did i say Tits? Tits!
Take out all references to fMRI scans/monitoring. What does the article say? Anything new? Why is this news?
Now, about that research paper... some interesting stuff about how the brain works. It'd be nice to know more about how the different brain systems and processes interact with each other and what implications this paper has. How do the processes interact? Do they compete or compliment or both? When and how? What effects does this have on our consciousness, i.e. our thought processes?
at work in human brains...
So the human brain runs Windows 8 ?
This is like asking "how many parallel processes exist in a stream of water. There are NO discrete "processes" in the human brain. It's a continuous, analog parallel computer. To characterize it as a "digital process" is exactly the same as comparing the human body to a locomotive engine because "obviously" both consume fuel, produce heat and generate motion.
thanks for the comment....i have one quibble: the Computability Function is redundant and doesn't explain or predict anything
it's a non-theory...it's like Epicycles...unneeded abstractions
Machines are made to execute instructions...end.
The whole "Turing Machine" is nothing more than a "what if" that has nothing insightful to contribute to understanding how to best make instrucitons for machines to follow
Thank you Dave Raggett