So here’s a thought… is it possible that the neocortex and it’s patterning abilities are actually a hi res version of what’s happening in much older low res structures in the brain? Could the amygdala be literally equivalent to an 8 bit processor which tries to assemble some low res meaning out of direct inputs from the senses and filtered (compressed) information from the cortex, to return a snap judgement to the cortex, or directly influence behavioural output if that judgement is weighted heavily enough. Obviously this low res output would have to be transformed in order to be used by the cortex…but it would explain why emotions are still involved in higher level thinking…I think. After an urgent event requiring a snap decision the cortex could store a memory of the emotional context to be incorporated into the overall memory of the event for comparison to the next similar event. Events not requiring the same swiftness of action would still be coloured by emotional state in order to determine whether new information could be hazardous or fun and thus chart a course for behavioural interaction.
I think it’s much more likely that the old brain structures do processing entirely differently than the neocortex. I don’t see them as very plastic learning systems but instead as heavily hardwired functional blocks that have limited capacity to perform.
Fair enough…I just figure once nature figures something out it usually just remixes it into a more complex form. I don’t think that I was suggesting that it is doing processing on an equivalent level nor would it need to. I think that the amygdala could simply be returning state information at various levels which then are assembled into a complex emotion by the cortex. The complex emotion could be represented as an SDR or incorporated into one, I’m not sure about the plumbing here, and this emotional component would colour future interactions not requiring an immediate unreasoned response. I think this could explain why some people black out or have no detailed recollection of extreme very short duration events, because the cortex was essentially bypassed and a purely reactive behaviour occurred. I am looking at this from the point of view that nature never really throws anything away, to use an analogy…biology built an apple computer, then it built an apple II, but nature never throws anything away so it wired the two together, then it built a Mac and so on…always wiring the new to the old and transforming the output of the old to be compatible with the new…the outputs from old systems are incorporated into more complex patterns and representations in these new systems…but the older systems still work for specific purposes so that functionality is preserved and the neocortex patterns it all.
For the example of retina -> visual cortex, the signal is passed through something but it’s hardly a single dedicated nerve fiber.
A post was split to a new topic: Are V1 transforms are pre-wired or learned?
I accidentally found a very interesting paper:
It says: “we report that a neuron can even learn a sequence of at least two, and probably more, accurately timed responses. A single cell is in a sense “programmable,” and can encode a temporal response pattern. This means that the nature of what a cell can learn is very different from the traditional view, and that the information storage capacity may be far greater”
As I can see the paper was published recently. I’m interested if anyone model those “programmable neurons” in their work?
I saw TM algorithm stores the active cells in the previous compute cycle, because neurons have memories?
Will the pyramidal neurons remain active for a while after being activated by feedforward inputs? how long?
Yes. If you read own papers, we always reference neuroscience work we base our theory upon. TM is based on the finding of different behavior based on different areas of dendritic activity.
Where can I find what you mentioned?
i read BAMI, it did not say the biological principle of TM.
I am reading the implementation of the TM algorithm,
Does these two lines mean that pyramidal neurons have memories?
The basic principle of HTM neurons is the the have a predictive state that extends one time-step forward. This predictive state is a form of memory.
but no evidence that pyramidal neurons have memories, right?
I think you’ll be hard-pressed to find a neuroscientist today that doesn’t think at some core level that our memories are stored somehow within the synapses between neurons. But they only make sense within large populations of neurons acting together.
yes, I always think that memory is stored on weights(synapses).
One thing I find really interesting that perhaps has something to do with intrinsic brain similarities and SDR differences is the apparent commonality of synesthesia manifestations in people who experience it.
Meaning a particular color evokes the same letter or number or a particular smell evokes the same sensation of touch in different synesthetes.
One short question from a newbie. For me predictive coding looks like some special case of imagination. Do predictive coding and imagination share the same machinery in the brain? How are they related?
Hold on a moment… when you say “predictive coding” what exactly are you talking about? Because that term can mean something different to different people.
OMG, yeah, it’s vague. Basically I mean the idea from “On intelligence” that our brain is constantly predicting what will happen next. Also I mean the mechanism in the temporal memory, when the neuron tries to predict what it’s input will be and it is used as a local learning rule.
Also, I would like to clarify what do I mean by “the special case”. As I understood from those predictive coding ideas, that prediction is the central part of intelligence and it’s absolutely neccessary for the brain to constantly predict the future. But I don’t feel that it’s the case. In some cases it doesn’t make sense to predict the future. For example, when everything around is very fuzzy and unstable. I guess sometimes people can choose not to predict the future. I see future prediction as some useful we often use, because we have imagination and imagination allows us to predict the future.
Also, as far as I understand in prediction coding theories feedback connections are used to transfer prediction errors, but I guess they are also used to transfer imagination to the lower areas
As far as I’m concerned, there are some cells in the body which is not recreated.
One of cells which is not recreated is neuron (brain cell)
Under my understanding from taking neuroscience lectures, human is born with 100 billions of neurons.
And I guess the reason that brain cell (neuron) is not recreated is to keep consistency for memory
because if neurons are replaced frequently like skin cell, that can mean not constant memory.
It’s known as one neuron is naturally dying every one second.
At 80 age, people are supposed to lose almost 3% of neurons.
And arithmetically losing 3% of neuron from age of born baby isn’t that huge, keeping constant memory through entire life.
But recent researches also note that some brain cells are regenerated mostly around hippocampus, or other areas of brain, using (maybe) adult stem cell.
Regenerating neuron around other parts from hippocampus is more slow.
Note that each cell of each organ has different period of regenerating.
Fast regenerating cell is observed in liver, skin, etc.
Anyway, human is born with full of neurons but connections are not prewired. Neuron and other neuron are dynamically connected via synapse connections.
Generally synapse connection happens in 2 spines of 2 neurons.
Spines are located on dendrites. But synapse can be formed in various regions like spine-neuron.
There are many dendrites to one neurons. One dendrite can have many spines. That’s the reason why there are a lot of connections in brain.
Anyway, connection changes dynamically via act of learning.
This kind of dynamical characteristic of brain is sometimes called plasticity of brain.