Brain Physiology, Cognition and Consciousness group: topic
This is a public discussion board
From Neuron to Memory System: How Memory Might Work
Graeme Smith
Friday, 12 June 2009 04:34 UTC
How does memory work?
We think we know, based on thousands of years of research into it, but there are still indications that we might be completely wrong. Scientists created computers with the best scientific approach to memory based on current scientific theories, in the 1940’s, and they got it wrong. Most neuroscientists today will tell you that the computer is nowhere close to a good model of memory, yet many of the attitudes that informed the decisions on how to make computer memories remain unchallenged. Our popular theories of how memory works were laid down in the middle ages, by people who thought memory was a fluid, and that the brain was a reservoir. Somehow they thought, we opened valves, and let the memory flow from the brain to where it was needed. I call this a demand model of memory because you demand the memory by opening the valves. In computing terms we have replaced the fluid theory of electricity but retained the idea of current, and at one time tubes which were replaced with transistors were called Valves. So computer memories could be seen to be operating in a demand memory model as well.
As Jerry A. Fodor said in The Mind Doesn’t Work That Way!:The scope and limitations of Computational Psychology Neither Phenomenal approaches such as Neural Networks nor strict computational approaches such as Truth Preserving Functions, seem likely to achieve a suitable computational model of consciousness.
As a researcher in Artificial Consciousness, my main thrust is to eventually get support for my own Artificial Consciousness to be built. However to get there, I had to start with a model of memory. I chose to hedge my bets, to begin with a Neural Model, and add Functional support to that model, where it was needed. Further I wasn’t going to limit myself to Truth Preserving Functions, where Soft computing would be more effective. This type of approach to consciousness is called a Hybrid approach. To start off, however I wanted a model of memory, and one of the things I had to do, to get there was overcome Fodors assurance that no-one had yet developed a phenomenal version of a demand memory.
Those who are stuck on William James idea that phenomenal systems can’t be built based on what we know about the brain, have problems with my use of the word phenomenal in the way that Jerry Fodor used it in his book. They like to look for exotic things like QM entanglement to explain the fact that some things just seem indivisible. I however think that there are indivisible elements in the memory but that it doesn’t matter because the mind doesn’t try to divide them, it finesses the system, without that effect.
Fodor’s claim is not without merit, phenomenal systems are definitely easier to build out of neural networks, than demand memories. However, David Marr defined a type of memory he thought might help explain the cerebral neocortex back in 1970, and by understanding what his work uncovered, and brushing off some of the unfortunate assumptions of his day, I think I have undercovered one of the hidden biases that is keeping us from understanding the way the brain, or at least the memory works. And that bias is the assumption that explicit (demand Model) memory, is the natural state of the memory system, and that anything that doesn’t fit the model must be an add on to the basic explicit memory model. So we expect Implicit Memory to somehow be an add-on to the essential demand memory, and we are wrong.
What Marr described in 1970 is a type of memory more basic than explicit memory when implemented in neurons. To computer guys like me, this seems counterintuitive. How can anything be more basic than a good demand memory, we can implement dynamic ram with a capacitor, a resistor, and a transistor, how much more basic can you get? But what we keep forgetting is that neurons don’t work like transistors, there is no technological overlap. Within the logic that makes neural networks work, demand memory is much more expensive to build than a simple content addressable memory. That is what Marr claimed his 4 layer CODON was, was a self-classifying content addressable memory.
To understand why this might be, we need a little theory. Although in the 80’s and 90’s the connectionist school was over-run by the Parallel Distributed Processing guys from the A.I. discipline, and so we have to take some of its theory as being deliberately misleading to steer people away from trying to understand real neural systems, The basics are fairly equivalent. A neuron is first of all a cell, it has to survive like other cells, by absorbing nutrients and getting rid of wastes. However at some point in the evolution of animals, neurons gave up some of their survival functions to a helper cell called a glial cell, and converted those functions over into mechanisms for transferring information between cells.
The Parallel Distributed Processing guys figured what was important to this communication was the firing of the cell. Sorry that is a misconception, firing just speeds the process of transferring information up, it is not the only mode of transfer, nor is it the most important one for understanding natural systems based on nerve cells. However because the PDP boys wrote the manuals for the industry, they got to tell 20 years of modelers what to think. As a result of mistakes like this, they set the neural modeling of Natural Neural Networks back, to the point where Gary Marcus, has clearly stated that he feels it necessary to do a hostile takeover of the connectionist school.
One approach that the PDP guys thought was dangerous was David Marr’s attempt to use probability Mathematics to define a circuit that was made up of heterogeneous neurons. Despite the fact that they did not have the ability to model heterogeneous groups of neurons, a direct attack on his claiming that the heterogeneous group he called a codon was a content addressable memory, was made pointing out that the model did not exibit wave-forms similar to the real cortex.
This attack, assumed that Marr’s circuit had to have frequency artifacts similar to the real cortex in order to suggest the role of content addressable memory for the cerebral neocortex. Well Marr was a pioneer in the field and he got some things wrong, and others even more wrong, but it was a seminal stage in the science, and what he got right is more important.
Marr was able to use probability mathematics to analyze heterogeneous networks of neurons and predict their function. Nobody before that even tried, and after the way he was treated few after him had the nerve to try again.
I am not going to start spouting probability equations, if only because I don’t follow his math. But the PDP guys called a neuron a processor that had input, process and output capabilities. I however look at it from a different perspective I see a neuron that has storage, processing, and transfer capabilities. And when I look at the 4 layer cortex model that Marr worked with, I also see a content addressable memory, but I think that perhaps he expected too much equivalency between neurons in the connections department, and was sadly dissapointed before his death in 1980. So I wonder at how much self-classification the system is doing.
If you look at neurons from my viewpoint you begin to see that the shape of the neuron has a logic, that memory neurons have lots of synapses, and opportunistic process growth, that processing neurons tend to have complex and often bushy or mossy dendrites, and that transport neurons tend to have virtually no division at the dendrite level and form long thin neurons.
If you look at the type of organs that are produced, Memory Neurons tend to form tissues, Processing Neurons tend to form globular organs, and transport neurons tend to form fibrous bundles. In other words the shape of the neuron, indicates it’s function, and the shape of the organ might indicate to some small extent what type of function it performs.
Looking at Marr’s 4 layer model, we see that the first layer is processing neurons, the second and third layer are memory neurons, and the fourth layer is dominated by processing neurons again. The whole structure is probably a memory role because it is a tissue.
Stimuli coming in at the first layer, are processed affecting their storage in the second and third layers, and the 4th layer probably is mostly control neurons, and in some cases inputs of senses that have been directed through the thalamus. These latter signals condition the general inhibitive environment caused by the Martinotti cells and thus reduce the inhibition at the first layer encouraging the layer 2/3 neurons to fire.
This simple memory circuit can be seen to be designed to respond to either signals from layer 4 or layer 1, and the first layer is used to train the second and third layers to respond to patterns of stimulus. Given a stimulus each pyramidal cell in layer 2/3 decides whether or not to fire, and this type of system is exactly what is needed for a content addressable memory.
It is also exactly wrong for an explicit memory suggesting that it must be the explicit memory that is the add on. Indeed when we look at the micro-architecture we see that a type of tissue called Allocortical tissue makes up the tissues at the bottom of the Sulcys and Divides of the brain. and since Sulcys and Divides are a later architectural configuration there might be an indication that the six layer tissue called isocortical tissue, that pushes the gyruses up away from the sulcys and divides, was a specialization of the brain.
-
Replies
Jump to resultsResults
-
Hi Hans
Ok, in order to understand my statement that the brain does not evenly distribute memory cells throughout the layers of cortex, let us start with the idea of a memory cell, If you remember I stated that a memory cell had certain characteristics to whit lots of synapses and opportunistic connections. I then went on to state that Pyramid Cells in layer 2/3 qualified for this role, in the Allocortical Model that I ascribe to Marr’s “A theory on the Cerebral Neocortex”, Proc. R Soc Lond B biology (1970) 176, 161-234
A second Look at the arrangement of cells can be found in Eccles "The Horizontal (Tangential Fibers)System of Laminae I of the Cerebral Neocortex (1983) Acta Morph hung 31, 261-284 However this version covers the full 6 layer Isocortical Tissue and deals with the arrangement of neurons into mini-columns and colums.
If you look at these maps, you will see that the pyramidal neurons tend to cluster in layers 2,3, 5, and 6. As I have already suggested layer 1 and 4 are consistant with processing neurons and are not therefore storage cells per se.
-
I think that it is very difficult and complex to substitute the nervous system that includes the memory, since this it works in an net of interactions difficult to control, it is too complex due to the quantity interactions that should use for alone construct a one word, one will never be able to create an artificial brain able to generate an independent memory, maybe if it is rotted to create robots with basic activities, but to think of creating or simulate a brain with everything, or that it implies that thinks in autonomous form it is impossible due to the quantity of connected elements, neurotransmitter, proteins, electric impulses that function in chain and one integrated interaction, as an everything, each millisecond to generate alone a word, is too many elements that to control and I see it impossible. Hopefully willing you can elucidate elementary aspects of the operation of the neuron, but if we think of the memory like such, it is difficult, there is too much elements that control, is as arriving to the one it limits of the expansion of the universe. I am not pessimist.
-
“[i]While the layer 4 cells are probably hooked up to the limbic and attention systems they really have little effect on the way that Allocortical Tissues Remember, the Cloud of unformatted data that they produce.[/i]”
The ‘lower’ areas set context where such is fundamental in understanding. First-time experience captures context such that later experiences, refinements in meaning etc come as particulars. Basal Ganglia has been mapped to context management and top-down control (pattern projection that ‘controls’, general over particular by setting the operational context) and the dynamics of the attention system in foreground/background management introduce particularisations as we zoom-in on details. This process of context (vague whole) and then text (details) is very energy conserving and brings out the ‘first impressions’ colourings of emotions contributing to a memory of some event and the refresh of such. It also brings out dynamics of ‘unconscious’ memories vs ‘conscious’ ones. For forground/background focus see such as:
Ivry, R.B., & Robertson, L.C.,(1998) “The Two Sides of Perception” MITP
Qiu FT, von der Heydt R. (2005)“Figure and ground in the visual cortex: v2 combines stereoscopic cues with gestalt rules” Neuron. 2005 Jul 7;47(1):155-66
These differences in classes of memories cover the many cortical levels we can map to different ‘brains’ covering basic stimulus/response of older levels (RAS and its single layer), through limbic and cingulate to the sophistication of the neocortex but also its highly particular focus resting on the layers of the past.
The identification, for example, of neocortical areas responsive to faces brings out a point that is supported by the underlying layers contributing context and ‘vague’ aspects of a memory that develop into a specfic in the form of a name – this dynamics brings out non-specific face recognition vs specific face recognition and so a more analogue focus (AM) translating into a digitial focus (FM) as we see in the overall brain dynamics and all the way down to the neuron and its AM/FM, wave/pulse, dynamic.
Since we can trace basic forms of categorisation at the linguistic levels to basic forms of categorisation at the neurological level so we have identification of the overall dynamic of information flow/management all the way ‘up’ from neuron to person to collectives and so from the single context, holistic, experience of a neuron-dependent life form, to the neural hierarchy that allows for symbol manipulation and the out-of-context communication skills we have where such includes classes of memories.
In that sense, since we have identified the INTENT of the brain, as in processing information as frequencies/wavelengths/amplitudes and the management of such through constructive/destructive wave interference patterns, so we are dealing with a dual system of local (pulse) and non-local (wave) dynamics that covers memory management.
In this context Han’s comment with regard to why focus on the brain becomes understandable in that the brain reflects millions of years of ad-hoc development (and so mindless, darwinian etc) that can be considered inefficient once we understand the overall methodology involved in information management.
Thus memories come in two basic flavours, pulsed, discrete, localised, and wave, continuous, non-localised. Your focus appears to emphasise the pulse perspective but the filtering of data that allows for stimulation of recall etc reflects more the position of translation of AM to FM (context pushes and as such encoding of memories/instincts/habits in dendrite bushes allows for recall and filtering of information as it does feedback to turn up/down the volume/contrast etc.)
Added to that is the recruitment issue where the encoding of memories in a specialist set of neurons is, when those neurons lose their interface (e.g. limb neurons in the sensory cortex) their recruitment by nearby neurons to increase THEIR bandwidth includes the transference of limb memories such that, if the recruitment is by face neurons, facial experiences (touch etc) can elicit ‘phantom limb’ experiences and so memory recall – indicating the filtering system model where the memories of the limb are still encoded in the neuron input areas and are filtering data that is now not limb associated but eliciting limb experiences (e.g. being touched).
Recruitment also brings out distributed memory where in a nuclei of neurons a memory can be distributed across all of the neurons as parts and so only recallable at the nuclei stimulus level, not the individual level.
The dynamics of difference/sameness management, and so anti-symmetry/symmetry dynamics covers the brain all the way down from consciousness to the neuron itself and allows for memories to be distributed at all scales besides a concentration of some specific within a level.
Your bottom-up approach (specific, layer X of the neocortex) covers a perspective of ‘all degrees of freedom are available, all is possible’ and excludes the top-down element that limits the degrees of freedom and imposes order, a pattern, that is then ‘meaningful’ and rememberable. IOW without BOTH approaches there is ‘nothing’ other than core neural patterns that dont relate to anything other than notions of ‘wholeness’, ‘partness’ etc etc
-
I hear you all, you want to jump in with both feet and get the memory system explained all at one shot. Using your own pet theory of course. Well you can’t deal with a complex system like that, you have to break it down into byte sized pieces. This thread is about how to build a memory system from neurons. So excuse me, if I start with neurons and build specific memory tissues before I jump full bore into each of your own pet theories. I assure you that despite dire predictions I can get further with this than has ever been done before.
But if the only reason you want to talk about this thread is to explicate on your own pet theory, I will continue to tell you that you are missing the point.So let me reiterate what I think is important. First the brain is made up of nerves. Secondly nerves do three things, they Store information, they process it, and they transfer it to other neurons. Knowing this we can classify neurons that do a specific role that concentrates on one of these three things, such as Storage Neurons that have lots of synapses and opportunistic connections, processing neurons that have bushy or mossy dendrites, and transport neurons that are mostly fiber and have minimal dendritic mass, and are long and thin. Using this, we can classify neurons as being most likely to be a particular type that plays a particular role. And, when we do, we find that the organs that have specific types of neurons in them according to this classification system seem to cluster in three types, Storage or Memory Neurons tend to form tissues, Processing Neurons tend to form globular organs, and transport neurons tend to form nerve fibers.
Using this classification system I have classified the neurons in Marr’s cerebral cortex model, and roughly confirmed his assumption, developed from his Probability Math that the 4 layer type of cortex, which I have found is called Allocortical Tissue creates a form of memory that might be Content Addressable Memory in support of Marr’s own contention. Further, I have determined the characteristics of such a memory and have noted that it has distinctly different characteristics than any previously published theory that I know of. To Whit it outputs a data cloud of redundant data, in reaction to every stimulus dumped into it, by the senses, and to other stimuli from internal sources. What is important about this data cloud, is that the opportunistic selection of which storage cells store which stimuli, and the characteristics of the Neural Network in which this memory is formed, make the data cloud unstructured enough that each individual organism that uses such a system must learn how to map meaning onto the output individually there is no common arrangement of data that could be used to isolate a particular memory when it is in this form. This indivisibility of the data-cloud form of memory has caused me to call it a Quale. The conclusion I have reached is that Allocortical Tissue produces a Qualar data cloud. Now please Attack these statements at will, but lets not push your own pet theory just yet.
-
Ottmar, finally a question that I feel qualified to answer! I had thought that you were going to bury me under stuff that is not germane to my theory.
Ok, I want you to understand that this is not my theory per se, I have simply bought the basic Neuroscience paradigm in this case.
First of all, yes the brain is made of cells.. There are predominately two types of cells in the brain, Neuron or Nerve cells, and Glial cells which can be further subdivided. According to Neuroscience it concentrates on the Neurons as being the active cells most involved in Cognition.
The Nerve cell looks completely different from every other cell in the brain. We think this is because it has been modified from the basic cellular structure in order to promote so called “Transfer Functions” that are related to communication between cells. Because this takes such a high priority, neurons or nerve cells, have to be supported by a helper cell, in order to stay alive, and this is one of the roles of Glial cells. Another type of glial cell promotes the growth of a white matter that when it surrounds a neurons Axonal process speeds the transfer of the signal down the Axon significantly.
Sorry the name of the white matter escapes me temporarily. But there is a definite disease caused by the breakdown of the white matter, It is a major debilitating disease suggesting that the brain relies very heavily even on this second type of glial cell for brain health. And for some reason, I can’t remember the name of the disease, even though a friend of mine has it.
Must be still half awake….. grumble.Ok. So yes, every cell communicates to some extent. In fact, in multicellular organisms growth is managed by communications between cells. The shape of an organ, is determined not only by DNA but also by the presence of certain “Factors” such as growth factors, etc. which regulate the growth. DNA is indeed long-term storage, of a life-time duration, in that we are given a single strand of DNA from each parent and we use those strands together to form the instructions that run our cells from birth to grave.
Like I said the main difference between this and a Nerve cell is that it uses often many of the same mechanisms that were used to regulate growth, and operate the cells, in order to promote transfer of information between cells. Think of this as opening new channels of communication that normal cells do not have, and you can’t go much wrong. These new channels operate using special chemicals called Neuro-transmitters, to communicate, and because the neurotransmitters are usually not used in cell to cell communication otherwise, the signals do not interfere with the growth of the other cells in the body but offer a unique opportunity to store and transfer data about the environment within the body.
Does this explication help?
-
Ottmar, there you go again. Please understand it is not that I am refusing to answer, so much as I want to develop my answer through a number of stages, Introducing memory circuits that each work a specific way. It would be premature to explicate the whole system until I have built it up a little from bits and pieces that make sense each in their own light.
At this point I am describing a specific Allocortical Circuit. Which I find, I may have erred in, by hewing too close to Marr’s original work. I am constantly attempting to fact check my work, so I asked a question on Science Solutions, and got informed that the 4 layer model, I have propounded doesn’t really exist, I am not sure what the scientist meant by Adjacent Layers, but it sounded like the fourth layer, never really descended, and that as a result, in Allocortical tissues many of the cells that would be in the fourth layer are interspersed with the pyramidal neurons in the 2nd and 3rd layers.
I haven’t confirmed this, but if it is true, then it supports my contention that Allocortical Tissue came before the re-routing of sensory data through the thalamus, and that Isocortical Tissue came at a similar time, because Isocortical tissues are both granular and agranular, (They have layer 4 or not) and so some of them developed after the rerouting of the sensory network.
An interesting question comes up, which layer constitutes the input connection in three layer allocortical tissue? are the connections to the second and third layers, or do they enter at the first layer.
To be more direct about your question, it is no doubt, that stimuli, affect the state of the neurons in the circuit I am propounding, and that this ability to detect particular stimuli, is important to memory, but I have not yet got to the point in the discussion where I deal with whether or not that storage affects the behavior of the organism. That determination can’t be made from the stage we are at right now, because there is no mechanism at this stage by which it could affect behavior. If anything we are looking at evidence that it could not directly affect behavior without that necessary step of interpretation that just hasn’t happened yet.
On the other hand, it is not that processing towards interpretation is not happening, just that we can’t isolate specific processing because there is no model to tell us where to look for a specific piece of information. All we have is a redundant data cloud with no apparent order to it. I will not leave you hanging for long, but I want to emphasize this point, because it is critical to understanding the next portion of my model.
-
Ottmar… good luck with bread, it takes a certain skill, to get it right.
You ask a good question about representation. So lets talk about that for a while Annette Karmiloff-Smith created the Representational Redescription Hypothesis, in which she presents the idea that the brain has to go through a process called redescription to get from what we call implicit to explicit memory.
The problem with implicit memory, is that you can’t recall it per se. It usually comes tacked onto a memory that you can recall, such as the color of your gandma’s bannister, tacked onto your memory of first sliding down the bannister as a child. This inarticulateness of implicit memory makes it hard to detect, but psychologists have shown that they can detect familiarity with a location even if you don’t consciously remember ever being there.
Because of this inarticulateness implicit memory is often thought of as an add-on to normal declarative memory. Annette said that instead, implicit memory comes first that people build up raw skill in a faculty long before it becomes possible to articulate about it.
What I have described so far, is a type of memory that has no method of being retrieved, and that voluntarily adds itself in the form of a data cloud whenever the stimulus is experienced whether or not the stimulus comes from within the organism or from outside the organism. I can be forgiven I hope from wondering whether I was also describing implicit memory. But in order to understand how it adds itself to explicit and declarative memory we need to understand explicit and declarative memory, and I haven’t got there yet.
So let us extract the hypothesis. I hypothesize that Allocortical (3 Layer) memory is pure implicit memory. In other words we know the mechanism whereby the implicit representation of memory might be formed, and I claim that it’s output is a disorganized data cloud and therefore Phenomenal, or Qualar in representation. This is what I call the Naive Cloud Hypothesis, because experience with the cloud of data after it has been interpreted might eventually lead to being able to make decisions without interpretation but in the naive state such as a babies brain no such distinctions can be made.
This means that the memory is operationally indivisible at its representation on output, and there is no mechanism in Allocortical tissue to allow any form of addressing below the cloud level itself.To test this I have proposed the following idea for an experimental protocol,
1. Pick a test of implicit memory based on olfactory and visual cues, and determine that both have implicit memory. Then pick a test of explicit memory based on olfactory and visual cues, and determine if both have explicit memory, and you might find, that Olfactory Memory has no explicit form. The primary dissociative test for implicit memory is primning and speed of response, the primary dissociative test for explicit memory is delayed response tests which should not show priming. Essentially if my theory is correct you should not be able to demonstrate olfactory delayed response memory, while you should be able to demonstrate it for visual cues.If this protocol is ever done, then it would either suppport or not my hypothesis that Allocortical Tissue is pure implicit memory. In other words any representation of memory that is done in the olfactory bulb, would be an implicit representation and not an explicit representation. This is not to say that there isn’t an internal logic to the representation, just not one that is discernable in a naive system. But it raises the question, can you actually recall the exact smell of your Grandma’s Apple Pie?
-
I have been busy. Just two thoughts for now. firstly, I am in favour of keeping this thread to a critique of Graeme’s model as originally suggested, because I think that will be enough for a while. Secondly, my grandmother did not make apple pie but my great aunt Ella made egg custard and her own Cumberland sausages in the 1950s and I can recall the smell of both with ease. Moreover, I can recall the smell of some substance X that I first encountered as an infant and have encountered maybe half a dozen times since, but not in the last ten years. I have never known what X is, or whether it is a single substance or a group. I think it is sometimes present in disinfectants but also sometimes occurs in other contexts. So it seems it cannot be an ‘add-on’ memory because I cannot precisely recall anything more about its context other than that every decade or so I recognise it, usually briefly without knowing its origin. I probably first encountered it in a salient context, but I do not remember what. I can ‘smell’ it at will and it is very pungent.
-
It is entirely plausible that my theory about allocortical tissue being implicit in nature is wrong, or I may have chosen an area of memory in the olfactory bulb, that has another architecture to achieve explicit memory with. I have a message in with an olfactoty scientist, who may know, and I am trying to get access to a psychologist that could help me polish up my protocol, so that it meets scientific standards.
Ottmar, I just came across an article from about 2005 that stated 5 types of representations. The bottom form was non-conceptual representation. I think that what I have described so far, might fit that bill.
The next step however is tricky, the question is how do you get conceptual representations from non-conceptual representation? How do you give a Qualar memory like I have described to result in isolated discrete memories that could be conceptual in nature? Mechanically it is easy, just define an address. But parametrically there are very few options as to how to map memories from the data-cloud that is not organized, into an organized map that you can demand memories from. In fact, I can only come up with one. isolate a specific data cloud, and test each address to see if it is part of the data for that cloud.
I welcome your discussion.
-
The computers never you obtain to bring near at least to like the abstract thought of the human brain works and of other animals, single obtain to design the elemnents but primitive of execution of certain abilities motorboats, but at level of the abstraccones it is impossible, there is too many variable inteconnect impossible of predict, the neuron possesses an infinite plasticity. I wait they don’t take me to bad.
Thank you for your attention,
Best regards,
Alejandro
Results
-
