5 ms·
Simulating a human brain itself was just the tag line, technically already accomplished by eg Izhikevich, TheVirtualBrain project, Eliasmith's SPAUN among other
by p1esk 7y ago
Simulating a human brain itself was just the tag line, technically already accomplished by eg Izhikevich, TheVirtualBrain project, Eliasmith's SPAUN among others.
No. No one has simulated a human brain. All those projects have simulated something that might vaguely resemble some of the brain's structures or operations, but we know so little about the whole thing that it was pretty much a pointless exercise.
HBP has been different from all of them because they attempted to do the true bottom up approach, not feasible without massive resources. As the article states, it turned out that even given the massive resources the task is just too difficult currently.
- derefr 7y agoI think you're ignoring the difference between a simulation (or model) and an emulation. A bottom-up approach produces an emulation. AFAIK, nobody has ever claimed to be trying to produce a human whole-brain emulation in the style of https://neurokernel.github.io/ https://neurokernel.github.io/. It's a problem that is obviously intractible right now.
- wuliwong 7y agoI don't think what you are saying is correct. Isn't an emulator something that takes the same inputs and produces the same outputs? I don't think "bottom-up" has anything to do with it.
- krastanov 7y agoI doubt most people reading the phrase "simulating a human brain" will make the distinction between simulating and emulating. The distinction is very technical and simply does not exist in the everyday use of the word "simulate".
- RandomTisk 7y agoAs a layman, I imagine the simulation would have to account for any one of the thousands of different compounds and chemicals in the brain where emulating it might be easier but a lot less useful, like testing drugs on mice versus humans, never knowing what real difference will make an important one.
- marmaduke 7y agoAnd yet the distinction is important as we are discussing a project on the very topic. I don't think anyone playing SimCity thinks there's a whole real city running on their CPU. And certainly no one has a heart attack if you say SimCity is a simulation of a whole city.
- Dylan16807 7y agoIf any model counts, then I can "simulate a human brain" as a sphere on a stick figure given five minutes in a physics sim. We need to have some standard, and I would say the bare minimum is modeling each neuron in an ultra-simple way along with the connections between them.
- Retric 7y agoDepending on what your doing modeling individual nerve cells may be pointless. For example, a stroke simulation may care a lot more about blood flow than communication between nerves.
- Dylan16807 7y agoIt's certainly a useful project but don't call your blood flow simulation or vague activity map "simulating a human brain".
- dTal 7y agoWhat if it turned out that neurons are highly redundant, and you can very effectively model network activity at a statistical level well above that of the individual neuron?
- slavik81 7y agoI've spent a few years studying biological modelling and simulation (of plants) but I've never heard of this distinction. Is it specific to neuroscience? What exactly is the "bottom" in a bottom-up approach?
- marmaduke 7y agoIt's a distinction that physicists have brought into the field, the bottom refers to the lowest level of detail, usually cellular but sometimes molecular.
- phs318u 7y ago> What exactly is the "bottom" in a bottom-up approach? Great question (to which I have no idea what the answer is). It seems to me that the various attempts are trying to simulate sets of emergent behaviours at different levels of the brain's "stack". I guess, one way of determining the success of a "lower level" simulation/emulation, is the emergence of known "higher level" behaviours?
- marmaduke 7y ago"every model is wrong, some are useful", models of whole brain exist and reproduce aspects of experimental data. My day job is building and evaluating such models; we have one on seizure propagation that is entering clinical trial.
- p1esk 7y agoWould you call a thermal model of a brain (3d heatmap) a "model of a whole brain"? The only "model of a whole brain" we have currently is the one of C. elegans, and even that's debatable because it does not model chemical processes which modulate neural activities, and therefore does not provide an accurate input/output mapping (in general).
- marmaduke 7y ago> thermal model of a brain No, it doesn't predict activity from parameters. > C. elegans, and even that's debatable because it does not model chemical processes "the map is not the territory". This sort of criticism isn't even scientific
- dTal 7y agoA thermal model sure does predict brain activity from parameters. I can say with high confidence that the brain will have very little activity at 0 degrees C.
- marmaduke 7y agoThat's a narrow range of identifiability with no relevance to the data we collect on a daily basis or the scientific questions being asked (beyond of course that it is a pedantic example)
- p1esk 7y agothe map is not the territory Not sure what you mean.
- yazanobeidi 7y agoYou should take another look at Eliasmith’s work. The neural engineering framework aims to be biologically plausible, with some assumptions made to simplify the math. Cellular processes are not the focus of the model. It however attempts to replicate these dynamics with simplified “components”. It’s one of the only, if not the only, neural models that is consistently supported by experimental data. Their website is https://nengo.ai https://nengo.ai and I am not affiliated with them just an alum that took his course.
- p1esk 7y agoI'm not against modeling aspects of the brain based on what we know, just don't call that "whole brain" simulation/model.
- marmaduke 7y agoWhat would you call SimCity if not a city simulator?
- AstralStorm 7y agoIt's a city simulator, not specific city simulator or even close. It is not accurate either as accuracy was not the main goal, but fun. Human brain is a very specific and complex brain which we do not understand - vastly less than even cities at any scale. And to produce any useful answers, the model has to be accurate to some degree - and we don't know how much. HBP indeed had no question which it tried to answer - it's more like climbing a mountain because it exists.
- p1esk 7y agoI have no problem with calling SimCity a “whole city” simulator. Why? Because we have a good functional model of a city. We understand its operation on any level. Not the case with the brain.