N.Wells
Posts: 1836 Joined: Oct. 2005
|
Quote (GaryGaulin @ May 21 2016,02:33) | The RAM based associative memory of a David Heiserman based model already links events and actions to each other in a sequential what comes after what way. The article seems to be explaining how neurons adapt to form the "addressing" part of the RAM circuit. I sent a link to Camp for their opinion but apparently they're not absolutely certain how it matches up to the Heiserman model either. My experiments indicate that what they are describing in the article is already accounted for in the model I have, more specifically requirement number 2, sensory addressed memory.
The authors are associated with Numenta, which is not "academia" but at least it qualifies as an academic corporation and one of the authors is listed as a "Doctorate". I'm not sure whether the journal would publish the work of someone with none of that at all. In either case the need for me to supply that sort of information is at the very least an embarrassing problem. The best I can do is list a High School diploma from a school I was transferred to that at the time almost lost its state accreditation due to students only being prepared for what are considered to be menial tasks like to pump gas, sand a car by hand, cook a meal, or run a cash register. Things have since changed for the better, but in either case self-learners who were ahead of their peers in science are punished by the higher education diploma mills and those who cater to them. If the work is not published by a major science journal that can be cited then it's OK to not give them any credit at all, especially when it's associated with a Theory of Intelligent Design. In a case like that it's easily justifiable to use the academic system to destroy any credibility they do have and I have good reasons to only want to empower those who do NOT advocate academic snobbery.
If I must supply appropriately cited work as opposed to "hobby" books and at work/home experience then those who must have it can kiss my ass instead. And entities like PBS who are likewise normally treated like they don't exist and are not given credit where due have good reasons for liking how things are turning out in the conference video and paper that very much turns the table on that situation. It's not often (or maybe never) that Dinosaur Train gets credit for excellent work in defining what a hypothesis is. Those who need long definitions that end up expecting the general public to stop using the word "theory" as is normal on US TV shows like the very popular program my wife loves named "Castle" and in real life are likewise part of the problem, but academia has a way of making sure that they don't even know it. More people than you realize have a low opinion of the academic snobbery that results in pompous demands and insults from university spokespeople who think they know what's best for us torch and pitchfork peasants. [/rant] |
Their model simulates biologically realistic, excitatory, cortical, pyramidal neurons, with typical neurons possessing active dendrites and thousands of ever-changing synapses, with different proximal, basal, and apical dendrites, which "closely matches known neuron anatomy and physiology". They propose a sequential, hierarchical process that allows neurons to integrate their inputs in a functionally meaningful way.
When stimulated, dendrites create NMDA spikes, which depolarize the dendritic membrane, thereby opening voltage-gated ion channels. The resulting influx of cations causes an increase in voltage. If the voltage increases past a certain threshold, that activates other voltage-gated channels thereby transmitting a current along the dendrite. When a sufficiently large and dense activity among apical dendrites, then a signal is sent to basal dendrites, and similarly if enough of those get activated then a signal is passed on to the apical dendrites, which generate an overall response by the cell. (From Palmer et al., 2014, Nature, http://www.nature.com/neuro......46.html "we found that isolated NMDA spikes typically occurred in multiple branches simultaneously and that sensory stimulation substantially increased their probability. Our results demonstrate that NMDA receptors have a vital role in coupling the tuft region of the layer 2/3 pyramidal neuron to the cell body, enhancing the effectiveness of layer 1 input."
Quote | [Back to Hawkins and Ahmad] Experimental results show that the coincident activation of 8–20 synapses in close spatial proximity on a dendrite will combine in a non-linear fashion and cause an NMDA dendritic spike (Larkum et al., 1999; Schiller et al., 2000; Schiller and Schiller, 2001; Major et al., 2013). |
An apical NMDA spike can cause a Ca2+ spike, which, if solitary, will depolarize the soma, but typically not enough to generate a somatic action potential. In turn, previously depolarized neurons emit a spike sooner than non-depolarized neurons.
Quote | If a section of the neuron's dendrite forms new synapses to just 10 of the 2000 active cells, and the threshold for generating an NMDA spike is 10, then the dendrite will detect the target pattern when all 10 synapses receive activation at the same time. |
Quote | The synapses recognizing a given pattern have to be co-located on a dendritic segment. If they lie within 40 μm of each other then as few as eight synapses are sufficient to create an NMDA spike (Major et al., 2008). If the synapses are spread out along the dendritic segment, then up to 20 synapses are needed (Major et al., 2013). |
This allows them to calculate the chance of false positive signals, which in turn allows them to calculate the efficieny of the system in distinguishing true signals from noise. Quote | In the above example, doubling the number of synapses and hence introducing a 50% noise tolerance, increases the chance of error to only 1.6 × 10^−18. |
I'm skipping over the significant portion of the paper that discusses how all this translates to learning rules for memorization of sequences, but "a network of standard linear or non-linear neurons with a simplified dendrite structure cannot easily implement these activation and learning rules."
This in some ways corresponds to your "confidence evaluation". In particular, you are going to get all excited that the authors talk about incrementing and decrementing. However, 1) the authors ground their simulation in actual underlying biological processes, 2) they simulate those processes, rather than just having variables with supposedly appropriate names, and 3) they document evidence about how those processes work rather than just making assertions. Most importantly, the authors instead use the term "coincidence detection", which is significantly less teleological and does not beg the conclusion of intelligence.
Beyond all that, no journal ever asks about authors' qualifications as a prerequisite to publication, or cites an author's affiliation (or lack thereof) as a reason not to publish. Reviewers may note those things as an explanation for why an author clearly has no idea what he or she is talking about, but that's a different matter. Among other things, retired geologists may continue to do research and to publish out of their homes, without listing university affiliations or degrees in their writings.
Numenta is a straightforward corporation (what is an "academic corporation"?). It is funded by investors in order to do research that is expected ultimately to yield profitable returns.
With the educational background you describe, you should hesitate to base any conclusions of your supposed understanding of how things operate (such as scientists being supposed to help cranks with their rubbish). You have demonstrated that you'd be best off with the starting assumption that whatever you think at first is likely to be wrong. (That's pretty much a good assumption to check every now and again in all of science, but in your case it's an all-the-time necessity.)
As a specific example of your instincts habitually being wrong, Dinosaur Train deserves no credit whatsoever for the definition of an hypothesis, as we discussed long ago when you first brought up the subject. First, credit goes to earlier people, such as Bacon, Chamberlin, and Platt. Second, Dinosaur Train is merely explaining it, not defining it. Third, as we have already discussed, their actual definition (an hypothesis is a testable idea) is not bad, and it works fine for kids, but scientific practice is a little more complicated. In particular, Dinosaur Train is conflating "hypothesis" as used in hypothesis testing with "hypothesis" as used in statistics, and is messing things up in the process. Fourth, their specific example sucks, because it is without significance or interest or explanation. In hypothesis testing in science, hypotheses (other than statistical ones) usually but not always contain significant elements of explanation, rather than just being declaratory statements of alternative realities that are about to be tested. Also, to be useful an hypothesis has to be testable (note the "Testable Predictions" section in the Hawkins and Ahmad paper). This stems from the longstanding use in logic of "hypothesis" as the antecedent of a proposition: it's A in "If A, then B".
Usage of "hypothesis" in statistics overlaps this, but is not identical: Null hypothesis: A is not different from B, at some level of significance; Alternate hypothesis: A is different from B at that level of significance. In this sense, we have indeed stated two mutually exclusive statements of possibility, and we test them against each other (as in Chamberlin-Platt hypothesis testing), but no level of explanation is included (like Dinosaur Train). It's simply two mutually exclusive statements (not just one, contrary to Dinosaur Train). Statisticians talk about testing these hypotheses, but this is probabilistic acceptance or rejection of the null hypothesis, rather than rejection by disproof. Dinosaur Train's example implies absolutist and positivistic rejection or acceptance, rather than either testing by falsifiable predictions or probabilistic evaluation.
Academia does not need reminding that America has a strong anti-intellectual / anti-elistist strain (George Wallace's attacks on "pointy-headed intellectuals"; Rick Santorum calling Obama a snob for wanting more Americans to go to college). Fortunately, science continually demonstrates its value to society by coming up with useful (and in many cases profitable) ideas. Rather like Numenta, and entirely unlike you, for example.
|