Could a human teacher enjoy teaching a neural network?

A place to discuss the science of computers and programs, from algorithms to computability.

Formal proofs preferred.

Moderators: phlip, Moderators General, Prelates

gbagcn2
Posts: 152
Joined: Sat Dec 27, 2008 10:36 pm UTC

Could a human teacher enjoy teaching a neural network?

Postby gbagcn2 » Fri Jul 30, 2010 12:33 am UTC

How complicated would a neural network have to be for a teacher to enjoy teaching it? It would probably have to be capable of things like asking questions and making mistakes but I don't think it would need full natural language understanding. Maybe it could happen over a decade before a computer passes the Turing test.

letterX
Posts: 535
Joined: Fri Feb 22, 2008 4:00 am UTC
Location: Ithaca, NY

Re: Could a human teacher enjoy teaching a neural network?

Postby letterX » Fri Jul 30, 2010 2:26 am UTC

I think the two threads you have open suggest that you don't really understand what a neural network is. They are not a magic box for pretending to be humans. They are an algorithm, just like any other, where bits go in and different bits go out. They may have some nice properties like being able to compute non-linear functions, but they don't automatically behave like brains just because they are modeled on neurons. So, the answer to your question is: incredibly complicated. And probably neural networks are not going to end up being the basis for the first sentient computers anyways (though this is pure conjecture, as true AI is a long ways off).

Where are you in your education? If you are at college and interested in these things, you should take a course in AI or machine learning. Even though neural networks aren't magic, the real subject is still really interesting.

kmatzen
Posts: 214
Joined: Thu Nov 15, 2007 2:55 pm UTC
Location: Ithaca, NY

Re: Could a human teacher enjoy teaching a neural network?

Postby kmatzen » Fri Jul 30, 2010 2:31 am UTC

I hope you realize that neural nets like Lieutenant Commander Data's brain don't really act like real neural nets. Your questions seem to indicate that you think this is what neural nets do.

gbagcn2
Posts: 152
Joined: Sat Dec 27, 2008 10:36 pm UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby gbagcn2 » Fri Jul 30, 2010 8:04 pm UTC

I prefer to learn things like this on my own so I have mainly been learning about them through the internet. Why do you believe neural networks won't be the basis for sentient computers?

squareroot
Posts: 548
Joined: Tue Jan 12, 2010 1:04 am UTC
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby squareroot » Sat Aug 07, 2010 7:40 am UTC

I like to think of neural networks like this - a neural network will take some input - say, a sound, an image, or a chunk of text - and you train it by telling it what output you want for each input; for instance, I would record myself saying "west" ten times, and tell the network to return the letters w-e-st. And then I might record myself saying "win", "test", "tune", "tin", "rest", "rune", "run", "nun", "nest", "tint", and "tent", and tell it how to spell each of those words.

Now, the way neural networks are built, they have neurons that look at specific parts of the input, and when you train the network, you're actually adjusting how neurons react to certain inputs. For instance, one neuron might "fire" and produce a signal (which gets passed to another neuron) if it recognizes a "ing" sound. But I would want it to recognize a "inge" sound, like binge. So the training would adjust the neuron so that it pays more attention to the part of the input near the nj sound, and less attention to the ng sound. Then, a nj sound would be more likely to cause it to fire than a ng sound. A ng sound might still contribute somewhat to making the neuron fire, because it possible there was bad sound quality. But a nj sound should be more important, and so the training changes the neuron's function so that it is more important.

So, the network learns one thing: if there's a "nj" sound, this one particular neuron should really pay attention to that, and should have a high likelihood of firing. What the neuron firing will do, that doesn't matter; but the training has showed through some techniques that are programmed into whatever software you're using that this neuron firing under those circumstances is good. Maybe that neuron firing ends up printing "inge" or "inj" somehow, which would be good. It's what we've trained it to do.

---

In regard to your original question, I can imagine a human teacher might get some pleasure out of teaching a neural network. In particular, if there were ten or twelve people teaching it at once, then progress would be more rapid and they might be happier to see it progressing as quickly as it does. I can imagine it wouldn't be extremely difficult to build a NN that could eventually learn to evaluate complex expressions, given lots of training - like sqrt(3)*sqrt(3+9)-1. But even then, dealing with place value would be difficult, and creating a suitable NN would require a lot of careful thought and prior knowledge on the programmers part on what types of syntax to deal with.

If a NN were to be used as the base for a computer, it would mean hundreds - if not thousands - of hours teaching your computer how to "compute". Sure, people might eventually program some knowledge in for it, and as people used these computers, they could update each other... but this could lead to unpredictable results. Suppose you tell your computer to save a file, and it tries to close a program. You get the little pop-up for whether or not to save. You say, "No, keep it open!" and all it recognizes is the "No", it might compute "'No' matches 'No, don't save'... okay, quit program.".

Even worse, in a distributed learning scenario, suppose a group of terrorists teach their computers to send out spam e-mail whenever they, say, mouse over the start button. If other peoples' computers learned that do, it would be terrible.

Of course, all the above situations won't even be really possible for a good long time. It would be ridiculous to make a computer that operates on a open, unpredictable system. And all this was written between midnight and 1 A.M., so don't trust me completely.
<signature content="" style="tag:html;" overused meta />
Good fucking job Will Yu, you found me - __ -

gorcee
Posts: 1501
Joined: Sun Jul 13, 2008 3:14 am UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby gorcee » Tue Aug 10, 2010 7:20 pm UTC

The basic concept of a neural network is this: an NN has an input layer, one or more hidden layers, and an output layer. The input layer takes a bunch of things as inputs. Because we program them on computers, these things are numbers. Sometimes those numbers have been derived from images, or sound, or data, or whatnot, but typically speaking, they're numbers.

Each input node passes its number to one or more nodes in subsequent layers. At these nodes, some mathematical operation is done, and a new number is produced. This goes on until you get to the output layer. The output is some number, typically one that represents some probability, but not necessarily. In the end, that's all a NN is.

Teaching an NN is basically training the network on known values, and running some algorithm to compute the weights at each node. For a polynomial NN, this would just be computing the coefficients that multiply the various values being fed into each node. These weights are optimized so that the error at the output nodes, trained against known data, is minimized.

So, could a human teacher enjoy that? I enjoy programming NNs for classification applications, and I have taught students before, so the answer is yes. But I suspect that if a teacher had no penchant for programming, nor a desire to study mathematics, then they would not enjoy it at all.

RabbitWho
Posts: 284
Joined: Sat May 01, 2010 10:16 pm UTC
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby RabbitWho » Wed Aug 11, 2010 12:45 am UTC

I remember being terribly disappointed that no matter how much I spoke Irish to my Furby, Meega, and kept him away from any other language interference he always ended up speaking English.

Spoiler:
I feel like Ralph Wiggam right now.

tuseroni
Posts: 124
Joined: Fri Jun 25, 2010 10:41 pm UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby tuseroni » Wed Aug 11, 2010 2:48 am UTC

ive seen good things from evolved neural networks. polyworld for instance uses a bunch of evolved neural networks, part of the fitness function was using the least amount of neurons (well, thats not quite true...neurons had a cost, they needed so much energy. the fitness function was more lineant. food showed up and the creatures went for it. fitness was determined by not dying, a good metre.)

the problem we have is we want to skip 4 billion years of evolution and go straight from scratch to human. if you did nothing else than get them to basic animal, i think we could take over from there.

of course anything GA is REALLY complicated involving a lot of things we dont understand. things you look at and say "why is that there?" "dont know but if you take it out the damn thing stops working" "but its not connected to anything" *shrugs* they lack the simplistic elegance that we as programmers like. things made by a GA are all ad-hoc and confusing...

but i do agree that the OP seems to be saying "neural network=AI" and its not. there are lots of other ways. but i for one still like the neural network.
Image

EduardoLeon
Posts: 111
Joined: Wed Sep 30, 2009 2:26 am UTC
Location: Lima, Perú
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby EduardoLeon » Thu Aug 19, 2010 3:55 am UTC

gbagcn2 wrote:How complicated would a neural network have to be for a teacher to enjoy teaching it?

That depends on the personal preferences of the human teacher. I tend to like crude machines more than I like my fellow humans (the cruder, the better), so a conceptually very simple neural network that can demonstrably (or experimentally) learn a varied enough set of (sorry!) mathematical functions would be more than good enough for me. Somebody expecting more human-like interaction has to wait until we have human-nervous-system-like networks, and the software to run it. If human engineering more often than not fails to build reliable software for our (relatively) simple businesses, do not expect too much from our ability to replicate millions of years worth of perfect non-stop work.

gbagcn2 wrote:It would probably have to be capable of things like asking questions and making mistakes but I don't think it would need full natural language understanding.

Well, here is the list of things about ourselves that we would have to replicate in order to build a computer that "behaves" like us:
- Either data about how we interact (at least, build sentences in human languages), or the means to collect that data (a highly parallel circuit testing our input against any "reasonable" grammar; and human language grammars are not know for being "computationally reasonable", so the circuits will be fairly complex).
- Either data about the things we know about our world (people, houses, food, cars, whatever), or the means to collect that data (good luck replicating human senses; we have enough problems with drivers for printing devices and webcams).
- Either data about the things it should be interested in learning (that depends on what the human trainer wants to teach it), or the means to collect that data (from the human trainer).
- The hardware and software to make all of this work. Reliably. Also, efficiently, because we do not want to run out of energy to continue living and functioning as a civilization, just to run a pet project for making a human-like machine.

Also, at some point, your human-like machine or neural network must have some things hardwired (which things?), because we cannot build something with an infinitely deep abstraction layer. And, from the experiment of designing and implementing systems based on or related to computation (one comes to my mind in particular: the C++ language and compilers), we should have learnt that, the more abstraction layers, the more "clever" we try to be, the more "meta" we go, the worst for our chances of success.

gbagcn2 wrote:Maybe it could happen over a decade before a computer passes the Turing test.

Nature built us in a process took millions of years, by building prototypes and killing those that were not fit. Do you expect to replicate all that work in ten or twenty years?
Gott weiß ich will kein Engel sein!

User avatar
Xanthir
My HERO!!!
Posts: 5410
Joined: Tue Feb 20, 2007 12:49 am UTC
Location: The Googleplex
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby Xanthir » Thu Aug 19, 2010 7:09 am UTC

EduardoLeon wrote:Nature built us in a process took millions of years, by building prototypes and killing those that were not fit. Do you expect to replicate all that work in ten or twenty years?

...yes?
(defun fibs (n &optional (a 1) (b 1)) (take n (unfold '+ a b)))

EduardoLeon
Posts: 111
Joined: Wed Sep 30, 2009 2:26 am UTC
Location: Lima, Perú
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby EduardoLeon » Thu Aug 19, 2010 8:22 am UTC

Xanthir wrote:
EduardoLeon wrote:Nature built us in a process took millions of years, by building prototypes and killing those that were not fit. Do you expect to replicate all that work in ten or twenty years?

...yes?

Okay, good luck building such a system.

Oh, and, after you are done building it, proving it correct, testing it, debugging it, documenting it... Good luck running it. I mean it, really.

And good luck paying the electricity (or whatever form of energy we might be using by then) bill.
Gott weiß ich will kein Engel sein!

gorcee
Posts: 1501
Joined: Sun Jul 13, 2008 3:14 am UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby gorcee » Thu Aug 19, 2010 5:05 pm UTC

EduardoLeon wrote:
Xanthir wrote:
EduardoLeon wrote:Nature built us in a process took millions of years, by building prototypes and killing those that were not fit. Do you expect to replicate all that work in ten or twenty years?

...yes?

Okay, good luck building such a system.

Oh, and, after you are done building it, proving it correct, testing it, debugging it, documenting it... Good luck running it. I mean it, really.

And good luck paying the electricity (or whatever form of energy we might be using by then) bill.


Killing a proto-elephant requires 2 years of gestation plus several years of maturing and fitness evaluation.

Killing a particular neural network parameter configuration takes between 0.01 and say, 20 seconds of Levenberg-Marquardt optimization and then about .0005 seconds of data evaluation.

There's kind of a big difference.

tuseroni
Posts: 124
Joined: Fri Jun 25, 2010 10:41 pm UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby tuseroni » Fri Aug 20, 2010 3:16 am UTC

yeah evolution doesnt take time, it takes generations. how long a generation lasts dictates the rate of evolution (evolution actually selects for dying soon after reproduction. old age is not beneficial because it slows the rate of evolution and in turn adaptability of an organism to dramatic change) so if a generation can be less than a second, you get a million generations in 1m*(<1) seconds (for 1 it would be 1 million seconds or ~11 days, for 500 milliseconds (more realistic even for a slow computer) 5 days. for 1 millisecond (possible for faster computers) about 2 seconds...)

evolution is MUCH faster in computers
Image

EduardoLeon
Posts: 111
Joined: Wed Sep 30, 2009 2:26 am UTC
Location: Lima, Perú
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby EduardoLeon » Fri Aug 20, 2010 4:27 am UTC

gorcee wrote:Killing a particular neural network parameter configuration takes between 0.01 and say, 20 seconds of Levenberg-Marquardt optimization and then about .0005 seconds of data evaluation.

There's kind of a big difference.

You must take into consideration the fact there is a huge complexity gap between artificial neural networks and nervous systems:

  1. Artificial neural networks are programmed to propagate signals in an orderly fashion, something that is ultimately enabled by the fact they are run in computers that have processor clocks. Nervous systems have no similar devices regulating the speed at which individual signals are sent. Thus, complexity arises not only from the signals themselves, but also from the order in which individual signals are sent.
  2. Artificial neural networks propagate digital signals in precise, predefined formats that do not allow infinite possible signals. This has the effect of limiting the amount of states any network might be in. I confess my ignorance on the details of how nervous systems work, but, if they are chaotic systems (in the mathematical sense, which seems plausible), and they allow a continuum of possible signals (which seems plausible as well), then complexity will arise in their behavior in ways that will not arise in artificial neural networks.

tuseroni wrote:evolution is MUCH faster in computers

Training artificial neural networks hardly qualifies as "evolving" them. The way neural networks "behave" is fundamentally the same after we train them. The only thing that changes is the set of numerical parameters that regulate the inner workings of each neuron.

To say an artificial neural network can actually evolve, it must be programmed to...
  1. Change its own "size," by creating and eliminating neurons as needed.
  2. Change its own "layout," by creating and eliminating connections between neurons as needed.
  3. Change its own "essence," by changing the way each individual neuron works. This one might require using some ideas from genetic algorithms; but this time, what we need to mutate is not a bit pattern, but an algorithm itself.
Gott weiß ich will kein Engel sein!

squareroot
Posts: 548
Joined: Tue Jan 12, 2010 1:04 am UTC
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby squareroot » Fri Aug 20, 2010 5:08 am UTC

I give you the NEAT algorithm - and it's many variants.

It does everything you just listed. :D Well, it can form new connections and add new neurons. And you don't really need to remove connections or neurons, you just set their weights to zero (but Progressive Pruning does actually remove some). And when you say "change its essence", I think that's pretty similar to giving each a neuron a set of functions (Gaussian, Sigmoid, Periodic) whose weights it could change too. CPPN and HyperNEAR, for instance, do something like that.
<signature content="" style="tag:html;" overused meta />
Good fucking job Will Yu, you found me - __ -

EduardoLeon
Posts: 111
Joined: Wed Sep 30, 2009 2:26 am UTC
Location: Lima, Perú
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby EduardoLeon » Fri Aug 20, 2010 5:35 am UTC

squareroot wrote:I give you the NEAT algorithm - and it's many variants.

It does everything you just listed. :D Well, it can form new connections and add new neurons. And you don't really need to remove connections or neurons, you just set their weights to zero (but Progressive Pruning does actually remove some). And when you say "change its essence", I think that's pretty similar to giving each a neuron a set of functions (Gaussian, Sigmoid, Periodic) whose weights it could change too. CPPN and HyperNEAR, for instance, do something like that.

Wow, pretty awesome. Can I adjust the order in which signals are sent, based on the parameters of each connection between neurons? If it can do that, I am downloading it right now.
Gott weiß ich will kein Engel sein!

User avatar
Xanthir
My HERO!!!
Posts: 5410
Joined: Tue Feb 20, 2007 12:49 am UTC
Location: The Googleplex
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby Xanthir » Fri Aug 20, 2010 1:32 pm UTC

EduardoLeon wrote:
Xanthir wrote:
EduardoLeon wrote:Nature built us in a process took millions of years, by building prototypes and killing those that were not fit. Do you expect to replicate all that work in ten or twenty years?

...yes?

Okay, good luck building such a system.

Oh, and, after you are done building it, proving it correct, testing it, debugging it, documenting it... Good luck running it. I mean it, really.

And good luck paying the electricity (or whatever form of energy we might be using by then) bill.

Dude, you're wrong on so many levels here.

So, first, evolution depends on generations, not years, as noted by a previous poster. We can run a generation *many* orders of magnitude faster than nature can. (Assuming base-10 magnitudes, 7-10 orders on current consumer hardware.)

Second, natural evolution wasn't aiming at intelligence. It spent ridiculous amounts of time fapping around with basic morphology before our proto-ancestors arose a few tens of millions of years ago. We'll be starting with a working model of neuronal structures, and using at least partially-artificial selection to push the process in directions we want.

Third, I know you struck it out, but proving it correct? Really? This is a brain we're talking about.
(defun fibs (n &optional (a 1) (b 1)) (take n (unfold '+ a b)))

EduardoLeon
Posts: 111
Joined: Wed Sep 30, 2009 2:26 am UTC
Location: Lima, Perú
Contact:

Re: Could a human teacher enjoy teaching a neural network?

Postby EduardoLeon » Fri Aug 20, 2010 7:40 pm UTC

Xanthir wrote:Dude, you're wrong on so many levels here.

So, first, evolution depends on generations, not years, as noted by a previous poster. We can run a generation *many* orders of magnitude faster than nature can. (Assuming base-10 magnitudes, 7-10 orders on current consumer hardware.)

Generations of potential human language parsers / real world semantics interpreters?

Xanthir wrote:Second, natural evolution wasn't aiming at intelligence. It spent ridiculous amounts of time fapping around with basic morphology before our proto-ancestors arose a few tens of millions of years ago.

Okay. (I really mean it.)

Xanthir wrote:We'll be starting with a working model of neuronal structures, and using at least partially-artificial selection to push the process in directions we want.

Yeah, but what guarantee do you have that artificial selection will not discard, say, 99% of the prototypes generated? 99.9% of the prototypes generated? 99.99999999% of the prototypes generated? One if the lessons learnt from genetic algorithms is that good fitness functions do not make up for bad mutation functions. How long will it take until we get a human language parser / real world semantics interpreter out of a model designed to perform glorified regressions?

Xanthir wrote:Third, I know you struck it out, but proving it correct? Really? This is a brain we're talking about.

Okay. (More like, well, whatever.)
Gott weiß ich will kein Engel sein!

User avatar
quintopia
Posts: 2906
Joined: Fri Nov 17, 2006 2:53 am UTC
Location: atlanta, ga

Re: Could a human teacher enjoy teaching a neural network?

Postby quintopia » Fri Aug 27, 2010 4:17 am UTC

a lot has been said in this thread, some of it approaching an answer to the OP's question and some of it wrong AND irrelevant, but I'm not going to say anything about either. Instead, I'm just going to suggest some reading to the OP who seems so eager to teach himself about AI.

1) The Engine of Reason, The Seat of the Soul by Paul Churchland
This book gives a lot of very compelling evidence toward the idea that a sufficiently complex recurrent neural network is sufficient to generate the systems of interaction and symbol manipulation that we term "consciousness" and "intelligence." Also, it comes with a stereoscope and lots of 3D pictures!
2) Gödel, Escher, Bach: an Eternal Golden Braid by Douglas Hofstadter
This book, among many other things, aims to show how a simple, mechanical formal system, like a network of fully deterministic neurons, can create emergent "informal" behavior, such as what we recognize as peculiarly human forms of intelligence: creativity, intuition, appreciation of beauty.
3) Women, Fire, and Dangerous Things by George Lakoff
Lakoff is a cognitive linguist, and this book practically defines what it means to be such: As a cognitivist, he examines the higher-level functioning of the brain, at the level of symbols/concepts and their organization, uninterested in how that organizational system is implemented in the brain. Yet, as a linguist, he does by analyzing and experimenting with human use of language. Despite the fact that he never speaks of neural networks, after reading the previous two books, it should be somewhat more evident how the metaphorically-connected semantic networks he describes could be implemented by such a neural substrate. And since human language is all tied up in the human condition (which Lakoff also demonstrates in this book), each being reflections of the other in the brain*, the ability to build a language knowledge engine on a neural network implies the ability to create abstract "intelligence" in the same network. Warning: This is a very long read, and requires a lot of focus, but I found almost every moment of it completely fascinating.

So my answer to the OP is this: read the above and the answer to your question will be more crystal clear than anything someone in this thread could speak of. Disclaimer: I am not responsible for any errors the above reading list may create in one's worldview. All theories about the actual functioning of the human brain and the nature of human intelligence are only that: unproven (yet uncontradicted) theories.

*Lakoff shows how language reflects the worldview of a human observer perceiving the world from a human perspective. The reverse reflection is the Sapir-Whorf hypothesis, which Lakoff does not argue for or against to my recollection.

User avatar
Solt
Posts: 1912
Joined: Tue Mar 27, 2007 5:08 am UTC
Location: California

Re: Could a human teacher enjoy teaching a neural network?

Postby Solt » Sat Aug 28, 2010 5:43 am UTC

Xanthir wrote:Second, natural evolution wasn't aiming at intelligence. It spent ridiculous amounts of time fapping around with basic morphology before our proto-ancestors arose a few tens of millions of years ago. We'll be starting with a working model of neuronal structures, and using at least partially-artificial selection to push the process in directions we want.


Yea, not to mention the time spent on biochemistry. I mean it had to evolve an immune system, how long does something like that take? We won't have to spent millenia twerking fur color or lose thousands of generations to disease or predators or extinction events. We'll include enough of a challenge to allow the networks to evolve new behavior but not as much out-of-their-control stuff as actually happened, which had almost no influence on their real neural networks. We won't even need to evolve all the brain functions, such as heart beat regulation or mental degeneration or recovery from traumatic injury.

EduardoLeon wrote:Artificial neural networks are programmed to propagate signals in an orderly fashion, something that is ultimately enabled by the fact they are run in computers that have processor clocks. Nervous systems have no similar devices regulating the speed at which individual signals are sent. Thus, complexity arises not only from the signals themselves, but also from the order in which individual signals are sent.


Actually brains do have such a regulation system- the length of the neuron itself. I don't know if our math can handle sorting out its influence, but it's totally doable in terms of simulation.

I believe the propagation time of neurons is on the order of milliseconds while a theoretical super-duper computer will be on the order of picoseconds.

EduardoLeon wrote:Artificial neural networks propagate digital signals in precise, predefined formats that do not allow infinite possible signals. This has the effect of limiting the amount of states any network might be in. I confess my ignorance on the details of how nervous systems work, but, if they are chaotic systems (in the mathematical sense, which seems plausible), and they allow a continuum of possible signals (which seems plausible as well), then complexity will arise in their behavior in ways that will not arise in artificial neural networks.


Can't we represent an analog signal to an arbitrary precision digitally by having an arbitrary long number to represent it? Besides, I don't think real neural networks work like that. Their strength comes from an extremely complex series of very simple operations- that is to say, each neuron does little more than modify the strength of the signal it receives and then distributes it. Modifying the network consists of nothing more than maintaining/creating/destroying these propagation paths. That alone is enough to create very complex emergent behavior from a structure as physically complex as the brain.



One place where this approach will lose out though is in energy. The brain requires 20 watts. We'll require a dedicated power plant to power an artificial one. Then of course there are the still unknown functions of the brain that may regulate it in ways we don't even know about yet. Do we even know how architectural organization is imposed onto the brain?
"Welding was faster, cheaper and, in theory,
produced a more reliable product. But sailors do
not float on theory, and the welded tankers had a
most annoying habit of splitting in two."
-J.W. Morris

VDOgamez
Posts: 122
Joined: Tue May 06, 2008 1:49 am UTC

Re: Could a human teacher enjoy teaching a neural network?

Postby VDOgamez » Wed Sep 01, 2010 5:50 pm UTC

I'd just like to say that I love teaching neural networks. Especially when they're ones that I'm implementing. But I'm kind of weird like that.


Return to “Computer Science”

Who is online

Users browsing this forum: No registered users and 6 guests