Dynamical Neuroscience

Pythagorean

I've just made a wiki article entry, and would be interested on input from professionals.

http://en.wikipedia.org/wiki/Dynamical_Neuroscience

thanks!

sshzp4

I am not going to critique anything specific but make this note: The moment you say 'dynamical' you mean a flow/difference mathematical mapping representation of a system. All comments about AI and the ANN neuron are absolutely irrelevant to 'dynamical neuroscience'. You don't study only 'learning and memory' type problems in 'dynamical neuroscience', you study the biological physics, chemical kinetics, electrochemistry, transport phenomena and emergent behavior of the systems as well.

So the article seems more suited to 'some differences between ANN and dynamical neuroscience'. I am going to recommend the article to be merged with some other article on dynamical systems and completely rewritten. Its got way too many facts muddled up and tries to make facts of personal conjectures.

Edit: The article is poorly written, ideas get thrown around everywhere. No structure and no clarity. Consider heavy revisions, and focus on improving single sections.

jackmell

I disagree with that. The moment I hear "dynamical" I think evolution equations and when it's coupled with neuroscience, I think "massively fed-back non-linear evolution equations". I think both AI and neural networks are extremely relevant to the idea of approaching the brain from the perspective of it's dynamic properties. Neural networks are non-linear dynamical systems and artificial intelligence, I am utterly convinced can emerge from such systems. Personally, my view is that problems in "dynamical neuroscience" are mathematical problems: how do we understand the non-linear dynamical properties of very complicated coupled systems of non-linear equations that we believe model the dynamics of neural assemblies.

I am happy to see more effort to approach the workings of the brain in terms of it's purely intrinsic dynamic properties and am optimistic that will lead us to a more complete understanding of mind, consciousness, and self-awareness.

sshzp4

Jack -- "I am happy to see more effort to approach the workings of the brain in terms of it's purely intrinsic dynamic properties and am optimistic that will lead us to a more complete understanding of mind, consciousness, and self-awareness."

You put too much confidence in NLD and ANeurons, my friend :)

Well my comment should be examined under the light of the wiki-author putting too much emphasis on a comparison between ANN based AI and dynamical neuroscience. It seems like a slap on the face of scientists who, say, model the neurochemistry dynamics and should be viewed as being dynamical neuroscientists as well. Without the experimentalists and without the biochemists working in the field, you wouldn't have physics to model and would be left with blind conjectures. So in a nutshell, my statement is "don't draw boundaries based on personal experience/opinions in the description of a field".

My observation of irrelevance comes from constructed implications in the article such as "Even in this day and age of lightning communication, Dynamical Neuroscience didn't even receive it's own wiki article until 2010". For a good reason, obviously. Since the field is nascent and borrows formalism better addressed under ANN mathematics, or dynamics and wikipedia is not the forum to engage in opinionated descriptions.

(You might like Seung and Lee's work at MIT on NMF algorithms. They show the statistical perspective of how signals can be processed through 'articulate' decomposition for learning to take place <emph> without <\emph> using NN's. Minsky is the reason I gave up on NNs ever being able to describe 'emergent learned behavior'. The concepts of consciousness and self-awareness are completely over rated and are of purely human interest, not engineering. The point here is: yes, the neural net model is cool; but only because it is easy to understand. They have been working on this since the 60's and have mostly failed to come up with anything other than math demos.)

So Jack, my global point is the wiki-author could be allowing personal biases to decide what a field is about or not. Is that right?

Pythagorean

sshzp, I welcome your citicism, but perhaps you could be a bit more constructive.

I believe I represented it this way throughout the article. Perhaps my mention of ANN was distracting and I should remove it. But you did see the reference to Izhikivech's Dynamical Systems in Neuroscience, no? This is an important point about thresholds vs. resonating.

I think I only bring that up in the beginning, but if you would like to point to the specific cases that bother you,

Well this is a vague criticism. Not very helpful, really.

Very vague, still. And don't you think it would be insulting to biologists for you to use them as an insult? I'm a physics graduate student. My only experience is with Morris Lecar networks. It's a work in progress. (I'm going to add a separate section on attractor networks).

Which was the point...

This is a good point. You might have read it before I added that part.



I figure since you're spending so much time defending your position, you way as well give more specific, constructive criticism :P

thank you!

jackmell

Well I don't want to go off-topic. Allow me however to reply to sshzp's comments:

I admire Pythagorean's efforts to make an effort at emphasizing the dynamics of the brain and did not feel it was poorly written. I do indeed have enormous confidence in non-linear dynamics and believe strongly it is the ultimate key in understanding how the brain works. However I am quite critical of the past 50 years of AI and agree they have failed miserably because their work has been based on models that are linear: transistors that are either "on" or "off", and the linear program. I do not believe the current implementation of neural networks will ever emerge artificial intelligence because they too are based on the current computer technology that is inherently linear and have always proposed that we will have to wait for a critical point in technology when someone creates a new qualitatively different device that is intrinsically non-linear. When these devices are then coupled in very complex ways to mimic the cortex, I have great faith this will lead to emergent properties that will be akin to real artificial intelligence.

Pythagorean

I really didn't mean to imply that, but if that's how you interpreted it then I need to rewrite the intro, because obviously it's made a terrible first impression on you.

My personal experience every time I tell people I work in Computational Neuroscience is they think I'm designing ANN, so I wanted to clear that misconception up.

I agree. I need to find a way to work this stuff in. I do have a section "beyond the neuron" to illustrate the point, but perhaps I need to make it clearer.

I think this is too strong of a statement. It's not the ultimate key, but it is definitely a fruitful pursuit.

jackmell

Allow me please to contribute something concrete:

W. Freeman's article, "How the brain makes chaos to make sense of the world" attempts to model the olfactory bulb by a system of non-linear delay differential equations:

http://sulcus.berkeley.edu/FreemanWW...ts/IC8/87.html

Also, Terrence Senjowski suggested strange attractors may have some part in memory formation in the brain. As you know strange attractors are a hallmark of non-linear dynamics. Terrence is co-author of "The Computational Brain". I do not have the reference where he makes this suggestion however.

One final note: I'm sure you're aware of the "Blue Brain" project where an IBM group is attempting to model the cortex. My understanding is that their work is centered on the (non-linear) Hodgkin-Huxley equations and have plans as I understand it, to begin incorporating "history" in the form of likewise non-linear integro-differential equations. We are aware that neurons exhibit such a "history" phenomenon: their present behaviour is dependent on their past behavior.

sshzp4

@Pythagorean
"I figure since you're spending so much time defending your position, you way as well give more specific, constructive criticism :P "

I will leave that task for your advisors :). I love to bark at all things (I get paid to do that), but am in general too lazy to start the process of a detailed review (unless I get paid to do that).

Since you are still a student, your effort is great for a school project. But as the reviewer of a professional technical review article, the article is shoddy. If I stumbled across that article while just surfing the web, I would have added a significant section with choice abuses. Reviewing and being reviewed are both hard processes, but I am sure you will find out. (So best get obdurated to that feeling from receiving vague comments that leave it up to you to find out the implications :D)

@Jack -- I do hope you are correct. But I will still quote you on the following "I do not believe the current implementation of neural networks will ever emerge artificial intelligence but have always proposed that we will have to wait for a critical point in technology when someone creates a new qualitatively different device that is intrinsically non-linear", and describe that as a mere conjecture or prediction, not the current state of truth. We have been trying ever since the perceptrons were conceived to create Bayesian NNs to process statistical data. That's still a hypothesis. My point is, right now statistics and trained classification seem better approaches to modeling intelligent behavior. However, this view can be debated based on the background of the observer. Most CS people will claim statistics is better but EE folks will call NNs a happier approach.

And Jack, remember NNs are conceptually more closer to statistics and classification, and are usually used as blackbox algorithms. NLD using perceptrons might lead to non-deterministic behavior (the feared counterpart of deterministic chaos), which can only be analyzed statistically. Anyway, quite irrelevant for the geometer's topic, but good to think about.

Sid

Pythagorean

It's not a school project, nor do my advisors know about it.


Point taken. It has changed a bit since I posted this.

And to be honest, you've actually given me a lot more than vague comments. You've giving me an idea of how certain types of people interpret my article, and that comes with identifying your own biases and unspoken assumptions. I actually have some work to do thanks to you and other, more gentle critics.

but I should sleep on it.

sshzp4

@ Jack -- Conjectures and hypothesis against a concrete theory constructed on experimental fact? You should use the term 'speculated' instead of 'suggested'. Wells speculated man could land on moon, gave him the privilege of being described as the progenitor of the idea for a long time. But 'gravity shutters' didn't work.

Speculation is merely the cautious way to claim that they said it first; if proven wrong they say it was 'mere speculation', otherwise its always an I told you so. Its a very dangerous form of academic fudgery.

EOT

Sid

Edit: Oh, I see the point of your last post. You were helping the geometer with references. I assumed it was an extension of your earlier comment. Anyway, never mind.

jackmell

I think it would be a good idea to have a neurophysiologist comment about the article (seriously). I'd be curious to know what they think about the article. I am a big believer in constructive criticism. No way you could swing that Sid right? Just asking that's all. And you're right, I've expressed my personal opinions about how the brain should be approached. I apologize for going off-topic and should have concentrated on the writing instead. Sides, I have an etouffee to make now.

sshzp4

@ Jack: Oh I could try swinging that! :) When you have been criticized enough (for everything from a misplaced punctuation mark to the presence of a hyphen in a misleading place) all criticisms are just indicators of the presence of issues that lead thoughts astray off the topic at hand. The more vague a critique is or the more destructive it is, the more the indication that you haven't been able to get the idea across. So the nature of a critique usually gives you an idea of where the issues with your authorship lie (Assuming of course that the reviewer grasps the language of the discourse and there is no conflict of interest).

apeiron

1) the page is all about neurons not brains, so should be called dynamic neuron science at most. Dynamical approaches to brains would cite the likes of Walter Freeman, Scott Kelso, Karl Friston, Stephen Grossberg, Paul Nunez, etc, etc.

2) the page is based on a fundamental misconception. Yes neurons/brains have a dynamic basis (like all biology), but what is important about them of course is the way they capture information. Talking about a purely "dynamic" approach is just wrong from the start (unless you have the explicit limited research ambition of studying the physiologic-dynamic aspects of their functioning).

Neural nets are a computational attempt to model what is going on (an informational basis to the information processing!). So there is room for a dynamical approach to information processing. Some people talk about hybrid disciplines like infodynamics.

But anyway, the page does not spell out where it sits on a spectrum of approaches (not that it is about "brain dynamics" as opposed to neuron physiology in the first place).

Pythagorean

As discussed with ssh, this is not the case, but it may be the result of bad communication. I may have changed the page to reflect this before you read it, but also note that I'm still in the process of adding sections beyond neurons. This is not, by any means, a complete page. I intend to add a holistic section.

Btw, the whole nervous system is of interest, not just the CNS.

This is the same case with neurobiology, neurophilosophy, and neurophysics, all of which are their own disciplines and have their own wiki articles. No one discipline accounts for all the aspects of the subject it studies. There are, however, many scientists, who are only trained in their field of analysis.

agreed...

Pythagorean

OK, major revisions, refined citations, added content. Please continue to point out overly speculative claims and suggest new sections or contet:

http://en.wikipedia.org/wiki/Dynamical_Neuroscience

Section to add, yet:
Applications (both medical and theoretical)
Chaos and nature
more cognitive content

Possibly this:
http://www.scholarpedia.org/article/...ausal_modeling

apeiron

It is looking better. But to focus things, what are you seeing as distinctive about "dynamical neuroscience" here?

To me, the central idea you want to articulate seems to be that neuroscientific approaches to explaining mind or cognitive function (the higher level stuff) has been based on a "too simple" model of the components. So a more accurate dynamical description of these components may serve as a better foundation for high level explanations.

If this is the case (I may just misread your intent) then it would be helpful to make a connection to the arguments that standard ANN modelling is too simplistic. And second, examples of modelling that makes use of more dynamical componentry.

The lurking thought when people stress dynamics is that there must be something big we have been missing by taking familiar linear, computational, atomistic approaches to modelling the neuron, and the brain. So if we go back to basics, we may finally unlock the secrets via some new dynamical principle.

I think this is true. But I don't personally think the secret exists "down in the neurons". I don't even think it exists in the collective behaviour of neurons or even, separately, at some whole brain level (as some like Nunez and Freeman sort of argued).

Instead, I believe these dynamical principles (actually they would be info-dynamical) would exist over all scales of neural organisation. They would be very general. Which is why I personally follow a systems science/theoretical biology/semiotics approach to modelling.

But anyway, the point I am trying to make is that you probably have a specific hypothesis about the reason for framing the research issues in the particular way you have. That is, we need to study neuron-level dynamics, component level dynamics, because somehow the secret we are missing can be found at this scale of mechanism. The existence of the page would make more sense if you spelt out this theoretical context.