Discussion Topic |
|
This thread has been locked |
jstan
climber
|
|
|
Dec 30, 2017 - 10:51am PT
|
A very interesting presentation:
https://numenta.com/papers-videos-and-more/
As a sufferer of CRS (Can't Remember Shit) this talk by Jeff Hawkins seems helpful to understanding my experience. In the past when remembering things I seemed to get a confirmation. Now remembering( I still often get the right answer) but now it feels like a guess. Jeff discounts the possibility the amplitude of neural spikes could serve as a confirmation. The demands for precision in this mode are too severe. He is very reluctant to abandon binary representations.
Jeff also discusses that neurological spiking involves a times series of spikes. I doubt that is the root cause because almost everything involves parallel pathways. How could two pathways always fail in synchrony? So I am left to accuse it to being a failure to adequately arrive at a good comparisons between two or more sparsely distributed representations. Sometimes, if I wait, I can get a confirmation which suggests timing or speed is the problem. Either that or decreased ability to compare SDR's quickly.
Time for MH2 to get back to work. Before I go I would like an answer to this.
Now that we can see black holes and neutron stars colliding, our most fascinating problem sits between our ears. New kinds of medical instrumentation will be needed.
|
|
Lennox
climber
in the land of the blind
|
|
|
Dec 30, 2017 - 11:53am PT
|
I think this article provides a good explanation of how the brain implements dynamic perception. It’s methods seem scientifically sound: it does not utilize vague or generic terms like “black box” or “transfer functions,” nor imprecise machine analogies of “sensors” and “actuators.” Instead it provides a sound model of what the neurons in a Biological Neural Network are doing.
Nonlinear Bayesian filtering and learning: a neuronal dynamics for perception
Anna Kutschireiter, Simone Carlo Surace, […]Jean-Pascal Pfister
Scientific Reports 7, Article number: 8722 (2017)
doi:10.1038/s41598-017-06519-y
Published online:
18 August 2017
Abstract
The robust estimation of dynamical hidden features, such as the position of prey, based on sensory inputs is one of the hallmarks of perception. This dynamical estimation can be rigorously formulated by nonlinear Bayesian filtering theory. Recent experimental and behavioral studies have shown that animals’ performance in many tasks is consistent with such a Bayesian statistical interpretation. However, it is presently unclear how a nonlinear Bayesian filter can be efficiently implemented in a network of neurons that satisfies some minimum constraints of biological plausibility. Here, we propose the Neural Particle Filter (NPF), a sampling-based nonlinear Bayesian filter, which does not rely on importance weights. We show that this filter can be interpreted as the neuronal dynamics of a recurrently connected rate-based neural network receiving feed-forward input from sensory neurons. Further, it captures properties of temporal and multi-sensory integration that are crucial for perception, and it allows for online parameter learning with a maximum likelihood approach. The NPF holds the promise to avoid the ‘curse of dimensionality’, and we demonstrate numerically its capability to outperform weighted particle filters in higher dimensions and when the number of particles is limited.
https://www.nature.com/articles/s41598-017-06519-y
|
|
High Fructose Corn Spirit
Gym climber
|
|
|
Dec 30, 2017 - 11:59am PT
|
Celestine Prophesy provides a tiny glimpse behind the curtain...
Tom, that's hilarious.
|
|
WBraun
climber
|
|
|
Dec 30, 2017 - 12:32pm PT
|
The gross materialists only study the hardware and NOT the mind as it is.
The mind is NOT gross material but subtle material.
The gross materialists are always bewildered and thus start guessing with their incomplete theories
always leaving out the most important component because they have no real clue what that really is ......
|
|
Largo
Sport climber
The Big Wide Open Face
|
|
|
Topic Author's Reply - Dec 30, 2017 - 02:09pm PT
|
A fun read per reality being continuous or discrete...
In a discrete space, say a square/rectangular tiled space, (for convenience) we start by constructing two sides of a triangle, each of 1 unit length . To traverse the hypotenuse from either point, we have to move one unit of length to the right (or left) and one unit of length down (or up).
Say AC is traversed in 2 steps, A-D, D-C we have a length of 2 units along AC in the tiled space.
Suppose we keep increasing the number of steps taken from A to C and decreasing the size of unit length, path along AC would look like this :n=1/16
Length along the zig zag path above AC is still larger than the length of hypotenuse by factor of √2, which was the same factor when we used a much larger unit of space and only 2 steps (n=2) to traverse along the hypotenuse!
This is essentially the Weyl's tile argument
the former result does not converge to the latter for arbitrary values of n, one can examine the percent difference between the two results: (n√2 - n)⁄n√2 = 1-1⁄√2. Since n cancels out, the two results never converge, even in the limit of large n.
This tells us that no matter how small a unit of length we take, not even an infinitesimal length, would even approximate the pythagorean theorem in a discrete space. It happens to be true because of the simple observation that you have to be able to travel across in space in any direction, which is, in this example, 1/2 to the right & 1/2 to the down (45°) simultaneously for a unit , and not a unit towards right then a unit to the down, which is what happens if we discretise length. For the pythagorean theorem to work, a fixed length measured along one direction must not vary when measured along another direction. This is known as isotropy of space, which is a property of the continuum. Discrete models with different structures other than rectangular can also be disproved using the same argument.
In a sense, this argument doesn't fall prey to unfalsifiable claims that there is discreteness, but beyond our abilities to experimentally observe. It doesn't matter how small the "grains" or "pixels" may be.
Take 3 sticks, two of them having a length of 1 metre and one of approx 1.414 metre, all of them measured along a common axis. Try to make a right triangle, if the hypotenuse falls well short of completing the triangle or after some rotation, extends beyond it, (heh) you're in a universe with discrete space.
About Time
Relativity itself only actually observes that there is “movement”, and “assumes” there is “time”.
If, I say, for instance “The bus arrives here at 9 o'clock,” I implicitly mean that the pointing of the small hand of my watch to 9 and the arrival of the bus are simultaneous events
This seems perfectly acceptable, unless you realise that we are comparing the co-ordinates( location) of one thing to a thing called “time”.
But in fact, the co-ordinates of one thing (a bus) are only compared to the coordinates of another thing ( the location of a rotating pointer, or the pulse in the circuit, in case of a digital clock ).
The point being, coordinates of space are used to measure time, so one could say they are really the same thing. If space is continuous, so is time.
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 03:06pm PT
|
Lennox,
However, it is presently unclear how a nonlinear Bayesian filter can be efficiently implemented in a network of neurons that satisfies some minimum constraints of biological plausibility.
Do the words it is presently unclear of the abstract mean anything to you as to the likely hood of having confirmed that BNN, in fact, do use NL Bayesian filters? They also question whether a BF that satisfies some minimum constraints biological possibility can be implemented in a living BNN.
They do say, Recent experimental and behavioral studies have shown that animals’ performance in many tasks is consistent with such a Bayesian statistical interpretation . But this far from having put the nail through the coffin.
Lenox,
it does not utilize vague or generic terms like “black box” or “transfer functions,” nor imprecise machine analogies of “sensors” and “actuators.
It must be one hell of a good article if it does not use these robotic engineering terms.
|
|
healyje
Trad climber
Portland, Oregon
|
|
|
Dec 30, 2017 - 03:12pm PT
|
Consider the standard definition of ...
Look, maybe change tactics. To date you've basically used the device of absurdly reducing complexity to a small philosophical box, logical meme or other trifle and then dismissing or discounting it. Your post on behavior is no different.
Again, with regard to behavior, it essentially defines life - no behavior, no life. And for all of you on the free will kick, a lot of behavior is deterministic via instinctual genetic encoding, a lot isn't. Driving into the latter to explore determinism is fun, but like a lot of things, is fairly irrelevant to life beyond an intellectual exercise so it doesn't get much of my attention.
Back to behavior and mind, both extant species and the evolutionary record show organisms evolved increasingly complex neural systems and sophisticated behaviors in lockstep synchronization - more physical complexity, more behavioral sophistication and that includes consciousness and mind. And through subconscious processing you can drive physical processes right up to the edge of consciousness / subjective mind.
So, for me the question is which is more likely: minds emerge from brains or new brains somehow channel a universal consciousness? When I look at the potential of 100 billion neurons and a 100 trillion synaptic gaps absent any remote idea of how those physical elements become receptive to a universal consciousness it's just not a hard call for me. Further, if there were a universal consciousness, why would meat be necessary at all; imbuing meat with mind seems, well, a mindlessly ludicrous and pointless exercise no matter how I look at it. But hey, that's just me.
P.S. Do give a shout if any of you are ever coming through PDX...
|
|
Lennox
climber
in the land of the blind
|
|
|
Dec 30, 2017 - 04:01pm PT
|
Dingus McGee,
You obviously have difficulty with reading comprehension. It is okay. A lot of people do. There is no shame in that. It is just something you should be aware of, and know that you need to put more effort into it just as I have to with math.
I suggest you re-read the abstract a couple times then read the whole article at least once, but also stop and re-read dense sections two or three times more as you go.
https://www.nature.com/articles/s41598-017-06519-y
hint - the authors are presenting a potential solution to the problems that make it “presently unclear.”
Since the very notion of biological plausibility is not well defined, the point here is not to argue that one specific dynamic is more plausible than another one, but rather to insist on the fact that the NPF does not depend on the importance weights and this provides a great advantage in terms of implementation (being neuronal or silicon-based).
|
|
healyje
Trad climber
Portland, Oregon
|
|
|
Dec 30, 2017 - 05:01pm PT
|
hint - the authors are presenting a potential solution...
I would say they are presenting interesting if simplistic algorithmic analogies for filtering / inference / sensory integration behaviors in organisms. And really, black boxing it and coming up with an analogy more or less corresponding to observed behavior is a great exercise in understanding ourselves even if we don't understand the wet implementation. It's also nice to see them give a nod to predation relative to these behaviors.
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 05:36pm PT
|
Lennox,
hint - the authors are presenting a potential solution to the problems that make it “presently unclear.
No hint needed. Did you miss that they have only established a statistical correlation? How many other leaning methods could establish a statistical correlation?
Correlations are not scientific evidence, let alone proof of how a mechanism works. There is a 100% correlation between when the sun is at noon and then Big Ben gongs 12 times. Does the sun at noon cause Big Ben to gong?
Lennox,
. It is just something you should be aware of, and know that you need to put more effort into it just as I have to with math.
That is a faulty comparison with you and me.
My reading comprehension of the abstract is fine. On the GRE I scored the same percentile in math as verbal. So there is about an equal likely hood that if I can do math I can also read. You admit you have no math skills and I can see you have less reading comprehension than me to think the quote you put forth substantiates anything.
|
|
Lennox
climber
in the land of the blind
|
|
|
Dec 30, 2017 - 06:06pm PT
|
Of course correlation does not equal causation. And I am not saying that this or that article is the final word on anything.
I do find it very interesting that this algorithm fits so well with what the brain has to work with and how it seems to use it, as do other things like Kalman filters and Markov chain Monte Carlo with other brain processes.
But of course papers like these do not move us any closer to understanding how perceptions, decisions, etc., emerge from the “wet” biochemical processes of neuron-neuron interactions. It will be interesting to see if future research can show how the “wet implementation” aligns with these or some other stochastic process model.
No argument that my mathematical ability is terrible. But at the risk of getting in too much more of a pissing match, I’m more than willing to put my experience with the English language up against your GRE score any day. There is a reason sycorax ridicules you; your verbal ability, vocabulary and reading comprehension are plainly quite mediocre.
I can see you have less reading comprehension than me to think the quote you put forth substantiates anything.
Reading comprehension quiz:
Where did I say the quote substantiates something?
Answer: Nowhere. The quote was meant to show that you were putting too much emphasis on the background information in the first half of the abstract and particularly in the words “it is presently unclear” as if those words were a disqualifier for everything that came after in the entire paper.
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 06:34pm PT
|
Lennox,
Reading comprehension quiz:
Where did I say the quote substantiates something?
Answer: Nowhere.
The quote was meant to show that you were putting too much emphasis ...
Are you kidding? Your wording, ...meant to show ... implies an attempt at substantiation. And you didn't catch your contradiction with all that English language skill?
Lennox,
But of course papers like these do not move us any closer to understanding how perceptions, decisions, etc., emerge from the “wet” biochemical processes of neuron-neuron interactions.
That is what I have said since your 1st post of these filter simulations.
|
|
Lennox
climber
in the land of the blind
|
|
|
Dec 30, 2017 - 06:41pm PT
|
So yes strictly speaking my quote did try to “substanitain” something. When I posted that I was thinking narrowly that you were claiming I thought my quote somehow proved the validity of the study. So score one for Dingus.
But you are still the Dunning-Kruger poster child.
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 06:48pm PT
|
Lennox,
So score one for Dingus
Thanks, Lennox. Some here never admit errors -- Werner never makes any.
|
|
Lennox
climber
in the land of the blind
|
|
|
Dec 30, 2017 - 07:06pm PT
|
Yeah, what I did was the same thing I was saying that you did.
Forest for the trees.
I like sparring sometimes, but I’ll stop with the insults, for now. It’s kind of pointless, and Werner should at least get some new material.
|
|
WBraun
climber
|
|
|
Dec 30, 2017 - 07:11pm PT
|
My material is ever fresh and full of fragrant blossom scents not like the dead sterile robotic gross materialists ......
:-)
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 07:16pm PT
|
Her posts are often littered with the same errors she complains of.
because Sycorax = DMT
|
|
Dingus McGee
Social climber
Where Safety trumps Leaving No Trace
|
|
|
Dec 30, 2017 - 07:27pm PT
|
Z brown,
No concordance?
at what granularity? MikeL and Largo have merged on a parallel mission. Werner & Paul have kept the same story. The rest of us are still confused.
I might choose the word, convergence? a la Michael Schermer
|
|
|
SuperTopo on the Web
|
Let us know!