2017

>>deep learning
its just statistics but non-linear.

thats it.

it wasnt until recently that computers got good enough to use it.

hardware hasnt really improved that much, we need to hit terahertz range before we can really make use of it, either that or have 1000 simple core processors.

>Basically an inverted pendulum
>Wheel lateral friction provides the righting force
>Combination of gyroscopic precession and rake/caster/trail cause the front wheel to steer into the lean angle
>Wheel lateral friction varies with steering angle
>Bicycle resists falling over
I saw a video of this contraption, and I'm not convinced their front wheel was genuinely non-castering.

Why would you know better than the Dutch?

Explain the creation of 3-dimensional room or rather room at all and time. We'll never know.

wings as in flapping wings? genuinely curious

best girl in her series

Here I'll explain consciousness for you:

"Consciousness" is a culmination of parts of the brain doing involuntary functions. With a social layer slapped on top, giving the illusion of thought and will. Your mind is just a collection of components of algorithmic-like functions that culminate together, with a useful sense of self and free thought put on top. Probably for reasons of social cohesion and communication in general. Think of it as a surfacing of functions, a useful buffer. How things change in your brain and this culmination works on a rigorous level, is extremely complicated and most likely something we can only interpret and replicate through excessive use of machine learning aid.

>not realizing the first and last items are the same

Don't EVER, post pictures of my wife without my permission.

Uh she's my wife...

Fpbp but I suppose there are some contrived bicycle-like questions that people don't know (or should care) about.

>why bicycles stay upright is something we shouldn't care about

Consciousness is easily explained by people hip to biosemiotics. It's just talking about sign relations, value, meaning, interpretation and code freak out the ignorant masses. So if I explain that consciousness is the result of the interpretation of the mind made possible by symbolic signs remembered by systems like animal neurosystems. That is being able to experience biosemiosis and remember it like animals can that is where minds become conscious. That's just hard to make sense of for those following the materialist dogma

>>consciousness
No. It is one of he most complex tings to ever exist.
>>wings
Well studied, to the point where accurate predictions can be made.
>>bicycles
Well studied, to the point where accurate predictions can be made.
>>quantum computing
They were BUILD BY HUMANS, although they are not perfectly well studied it is known EXACTLY how they work.
>>gravity
Well studied, to the point where accurate predictions can be made.
>>deep learning
Even better studied then quantum computers. It is also a human invention and wasn't created by chance.

underrated post

>What are bikes
>WE JUST DON'T KNOW

Youre dumb

>Well studied, to the point where accurate predictions can be made.
Bruteforcing a system with computers doesn't mean you understand it well.

Scientists still can't explain why Japanese people want to fuck their Imoutos so badly

It demonstrates you understand it enough to make it your little bitch.

In the end that's all that matters.

Otaku don't usually go outside so the only opposite sex interaction they have is with their mother and sister. I'll let you figure the rest out.

?
Nice ad hominem, great argument.
It does.
Have you even taken ANY course in engineering or physics?
If the models engineers have provide no understanding then we understand LITERALLY nothing.

But because it works very well and the models are reasonably accurate we understand reality.

Also planes AND bicycles existed BEFORE computers. And even then we had very good understanding (and good models) of how these things work.

>And even then we had very good understanding (and good models) of how these things work.
Show me a pre-computer good model of a bicycle's mechanics.

OP BTFO

Ok I'm just stop you there, magnets are fucking magic.

>2017
>OP still can't find
>a gf

Dude. You have access to the internet.
Google it.

en.m.wikipedia.org/wiki/Bicycle_and_motorcycle_dynamics

Read the history entry.

Also note that models are NOT computer generated.

Again. Take a course in engineering and you will see that models are pure mathematics.

The ONLY thing you do on a computer is "plugging in" the values, because it is hard to by hand.

All of that is on computers/the internet though. I said pre-computers

>2017
>Veeky Forums can't grasp GF theory

The history section supports my assertion.
It took until 1970 for someone to point out that gyroscopic effects were not necessary and all later research involved computers.

I don't understand what you are saying.

Models have ABSOLUTELY NOTHING to do with computers.
Computers never helped creating models.

Every explanation you hear about bicycles has NOTHING to do with a computer.

If you read what I pointed you to you would have also seen that these models are far older then any computer.

Retard.

Your ignorance is not an argument.

No they're not.

From the cornell documents.
>More recently, emphasis has been placed on the development of highly detailed nonlinear mathematical models and digital computer simulations of vehicles which have permitted the investigation of aspects of performance which could not previously be perforrned with confidence. In conjunction with these developments, computer graphics techniques have been adapted to provide an integrated conception of the overall rnotion behavior of the vehicle.
Read your sources kiddo.

You obviously have no Idea what you are talking about.

This really annoys me even more.
There is a difference between a MODEL something which is a mathematical concept and SIMULATION which is using the model to get results.
Simulations run on a computer MODELS do not.
It said that the simulation couldn't be done good enough. NOT that the model was inadequate.

You are also starting to really contradict yourself. Everything you say is a result of your ignorance and lack of knowledge.

>bicycles
OH MY GOD, THIS Veeky Forums RETARDS IS SO STUPID! I cant believe how could you be so retarded! Thats a fucking simple precession! How stupid you supposed to be to forgot about it? Look at fucking gyroscope! youtube.com/watch?v=GeyDf4ooPdo

Uhh bullshit. he is clearly just a strong guy that can manage to hold it with one hand. Then he pretends to spin

>highly detailed nonlinear mathematical models and digital computer simulations
AKA Bruteforcing a system with computers.

Muh gyroscope!

Stop contradicting yourself.

It even states that the model has NOTHING to do with the computer. ONLY the simulation has.

A model gives you the understanding.

The simulation is needed when you want to see what the model predicts for SPECIFIC situation.

You don't know the basics of how any science works and you are still arguing about it.

Just to summarize. The model gives you understanding, the simulation gives you results for specific situations.
Exactly ONE of these things can be done by a computer.

Now read a book and stop being so ignorant.

No, he just a new generation of human-like androids.

He did say the thing weighs 40lbs. That's really not that heavy. I'm fairly weak, and i'm pretty positive i can lift 40lbs over my head with one arm. If he works out a little, he could probably swing it around like that. I'm not covinced unless i can try it out myself, or i see someone trustworthy do it.

>A model gives you the understanding.
The model is too complicated for that.
Imagine I had a magic super computer and I could model every atom of the bicycle, I wouldn't really understand it.
The complexity of the bicycle is inaccessible to current techniques. No shame in admitting it.

It's pretty simple

You really dont get it, do you?

A model is not really that complicated. (If you have a basic understanding of mathematics and the subject matter).

It can EASILY answer question like: "Why does a bicycle usually stay upright?"
Read about these models and try to understand them.

But when you want to ask harder questions like "What happens when I drive a bike with some layout on a specific road with a certain speed?"
You need a SIMULATION. That is the difference.

>Imagine I had a magic super computer and I could model every atom of the bicycle, I wouldn't really understand it.
But that has nothing to do with how you model a bike.
This is just a false equivalence. Stop it.

>The complexity of the bicycle is inaccessible to current techniques. No shame in admitting it.
Wrong.
YOU have no idea how it works thats why YOU believe it has no solution.

You really are burying yourself in fallacies and contradictions. YOUR OWN IGNORANCE IS YOUR ONLY ARGUMENT.

No you cant. Watch on 1:33
youtube.com/watch?v=tLMpdBjA2SU&

this

>"Why does a bicycle usually stay upright?"
it doesn't.

>It can EASILY answer question like: "Why does a bicycle usually stay upright?"
Why do they say otherwise?

This is written so fucking nicely and clearly. I hope to one day be able to explain things this eloquently.

Deal with it. This stupid Veeky Forums dumbfucks are just too retarded to understand the simplest things. There is no use to explain anything to them.

Uhhh, you are on Veeky Forums too though, you dumbfuck. Now you'll probably claim you come from Reddit or 9gag.
protip: that's even worse

youtube.com/watch?v=YdtE3aIUhbU
So how does it steer into the fall if there's no castering?

>Nice ad hominem, great argument.
welp now I agree with user. you gotta be pretty fuckin dumb to think that being insulted for saying dumb shit is the same as fallacious ad hominem arguments.

but you literally just described a materialistic dogma... unless you're not equating the structure/ symbolism corresponding with conscious events with the actual conscious aspect of said events... in which case you've not explained consciousness itself but rather necessary correlating physical events (which is fuckin great and all don't get me wrong).

To be honest, I could've done better with that reply, now that I have time to be critical of it. There's a lot you can't say with this character limit, and there's a lot you don't say when you post in a fervor while trying to fit it all in one post.

If it's any consolation you now seem like a dick

lel

I came from youtube.com you dickhead kid. Do you really think there is only 3 webpage on whole internet? Thats why you are too stupid to understand how mechanism like bicycle is working.

This

No, they can explain deep learning.
Its a meme, neural net operations are just sets of matrices.
Just no human would WANT to backtrack and analyze the derivation shit from features

Thats why theres visualization software now

>Do you really think there is only 3 webpage on whole internet?
>Came from YouTube
My fuckin sides

Also, who the fuck writes .com? Given that and your "dickhead kid" spit I take it you're something of a geezer

>Thats why theres visualization software now
Like what?

geez u r so dumbfuck man, kys.

Bicycle therefore bicycle?
1. Bicycle. |- Bicycle.
___2. Bicycle (Given)
___3. Bicycle (2)
4. Bicycle > Bicycle (2-3 Conditional Proof)

By proof: Bicycle, therefore Bicycle.

user this might be a long shot but you seem smart. Have you read Peter Watts' Blindsight? If so, other than the "vampires", how much of the ideas in there are popsci bullshit and how much are fairly accurate?
please help out a curious brainlet

> 1000 simple core processors
That's literally what a GPU is. And I do mean literally.
99.9% of deep learning is done on GPUs because they're like 100x faster than CPUs.
And terahertz would generate so much heat it wouldn't be practical whatsoever. That's why nobody goes above 5 GHz these days, because of the heat output.

Deep learning is pretty simple actually.
It's essentially just nonlinear regression through an indefinite amount of dimensions (ie 1024D input and 10D output for classifying 32x32 handwritten images of letters, for instance)
Training is updating the coefficients through gradient descent.

playground.tensorflow.org/

>ie 1024D input and 10D output for classifying 32x32 handwritten images of letters, for instance)
Elaborate.

you pedal to turn a power sprocket which transfers the torque to a wheel sprocket which then turns the wheel and causes the bicycle to move.

you steer from side to side to prevent the bike from falling over.

It's good as it is. Curious to see in what ways you felt you could have done better.

>There's no point in trying to explain away something that is wholly contingent with more than one scale (i.e. quark to apple to planet). It's too wide and fluid a situation for there to be some ultimately specific reason or rationale to pin to it, that would "mean" anything to human beings.

This point ( as well as he following analogy you used) sum up life.

I'm not convinced. Their contraption has the wheels at different heights. When steering, the force due to the gyroscopic effect might be the same by both wheels, but because of their different height the moment these forces exert on the bicycle body itself will still be different and therefore there will still be a net gyroscopic moment on the bicycle. So, gyroscopic effects are not ruled out.

Is a pogostick a trivial bicycle?

Watts cited sources for all of the concepts in the book.

> 2017 and deep learning are the same thing
Woah

playground.tensorflow.org

CUDNN from CUDA also allows you to interface with each individual neuron specifically to do shit like this

Also a few other frameworks i can't recall atm

What is it that perceives the illusion you speak of?

What about

Well I was including information as outside what is considered material.
I was explaining the correlating physical events but wouldn't it follow that consciousness is the subjective experience of the mind?

Neruoscientist here, confirming the hard no on consciousness. Studying consciousness is really just studying what it is like to have an experience. Some things (the psychophysics of sensations, properties that distinguish consciousness from unconsciousness) are well known, but there isn't a clear path to either explanation or prediction of conscious states, which among other things will require a causal, neurobiological explanation for "that thing you experience when you experience a thing." It remains unclear how the causal (neural) generators of consciousness and the metal experience of consciousness are related, and how (much less why) any specific neural event should give rise to a specific conscious experience. Indeed, we do not even have a scientific understanding of how (or why) neural events give rise to any conscious experience currently.

My main peeve is the register, maybe. I could have chosen better words, could've made the sentences flow better. But you might be right, those two sentences convey more or less what I wanted to convey. I suppose I'd want to clarify more by suggesting that the "ultimately specific reason" would be something like "x is y because z, and because I can only perceive z, z is the only reason y leads to x, bar none. As if to say, the only concepts you can perceive or understand are right, because you can perceive or understand them. That, it isn't possible for a slew of other variables to come into play, every snapshot of a moment, and that one rigid model could never hope to account for all of these variables- unless they are all present and accounted for in these snapshots of a moment, the moment's width depending on how particular you are regarding the whole operation. Human brains will never be able to match the speed of a bunch of cores strung together... because it's like expecting a basic dollar-store calculator to compete against, well, a supercomputer.

Some people don't understand magnets. It doesn't (shouldn't) mean they're magic, and it certainly doesn't (shouldn't) mean there's only one way to explain the mechanics behind them while still being some degree of accurate.

I feel like I could've spent maybe 2 or 3 more posts breaking down just what it is the human brain does (according to various sources), and what it is the most "common form" of deep learning would do (again, sources). I could have then went into a tangent about the early metaphysical "errors" through history (to me at least), revolving around epistemology and the understanding of nature, and thus the universe. Aristotle, the Heliocentric theory, all that. What knowledge is commonly known to be understood as, what it used to be considered to be understood as, and what might be wrong with not ratifying those concepts to better fit the reality of the situation as it appears to be. If I had the energy, I'd break down what each chosen example through history meant to convey, too. You know, all the different ideologies regarding existence, and logic, and whatnot. You've got your fundamentalists, your functionalists...

Then I'd attempt to break down language and the brain at the same time, step by step, and paint a cohesive enough picture of how spectacular, amazing, and relatively simple they are, and point out the numerous categorical errors that are sometimes made. Exactly enough, break down what you use to understand, to understand that what you use is more and less than what most people understand it to be. Obviously, there might be backlash, because it should be evident to anyone that your concept of your brain and your "innate" understanding of language is quite exactly the center of what will go for your beliefs. How do you come to terms with being "wrong" about something as core to your understand as, you? About how you work? About how things work? A lot of people don't.

Then I'd try to break down the underpinnings of deep learning in the same steps it took to break down the above, draw the parallels, and then make the discrepancies as clear as possible. I'd try to make clear just how... terribly unfortunate it is to believe that an explanation of how logic can be represented through and by something that is not alive or human, is not an explanation, because it doesn't confort this idea that the words I type now are true knowledge, and that "3,7,9,18", while not meaning anything in terms of the words I type now, also doesn't mean anything in any other form given any other context (provided the context even needs to supplement/explain away the number sequence to allow it to mean something). The fact that the number sequence should seem so alien, abstract, and pointless, is almost exactly the point I was trying to drive, and I'm just left wondering if there's a better way to bridge that gap, to paint a picture that seriously doesn't/shouldn't exist besides to illustrate the point. That logic is, as we can see it to be, something not innately inside us. That knowledge may as well be a tool, rather than the driving force behind the world. That what we come up with, isn't the end of it all, not even close, and that while we may not "understand" something a machine spits out, it does not mean that (all) the correlations it can manage to create, the models it manages to generate, the conclusions it may spit out, the actions it may suggest, are meaningless or pointless for humans to work with.

I mean, a great example of not wanting to accept everything coming out of a machine is when it tells you that there's a correlation between the temperature of some sticks of butter in a cellar, and copy paper stocks, when you feed it data sets from the LHC. Either there's something fantastically amazing going on there, or it's... well, not relevant to trying to determine if supersymmetry is valid.

In fact, here. That last bit, that last example that tries to illustrate the mistake (?) of assuming that deep learning is flawless, and that you should abandon everything you know- or the inverse. I can do something with that.

One thing I wanted to bring up is "correlation is not causation". Often, we go into things with a premise, or premises. A hypothesis, or hypotheses. Something does this, so something should do this, if we find that in our experiments or our calculations that something does or becomes this, we have confirmed... our suspicions. Our pre-determined notion, our assumption, what we're looking to disprove- or in the event that we cannot cause any other outcome, prove. The problem is, we're controlling for specific variables for specific outcomes, and often we sometimes try to apply those rigid conditions to something as granular as "quark, to apple, to a planet". There could be things in such a setting that our model may, as a consequence of being of our time, fails to account for. There will be a limit. And that's fine, if we really are only concerned with finding "this" specific thing, whatever that is.

And that should seem like the way we should do things for the most part. The problem is, again, how many times that's made us look the fool in the past. The Armillary Sphere in Italy is just an example. You'd get some amazing predictions out if it- but of course, the Sun doesn't orbit the Earth. But you wouldn't need to know that to, again, use it's predictions correctly. Specific, or multiscalar? I don't mean to glorify an extreme of one or the other (induction versus deduction), so much as illustrate how the operation of a "human" process, and that of a machine, differ. In turn, how they should be treated. You can't expect a made hypothesis or theory when the only discernible, human-brain-good-mean thing we as talking meat are given to parse from the machine, is outcomes, or straight answers. There is none (usually).

I accept that, even though the implication was I wish I could have came off as less of a dick about it.

How wings and bicycles work are very well understood phenomena

t. Aero engineer

Why would an aerospace engineer know about bicycles?

In one sense deep learning is as simple as you described it.

But in another sense it is incredibly complex, and we really don't understand whats happening. Why do our optimization algorithms work, when we have no guarantees of convexity of our loss functions?

Why do the solutions arrived at by optimization seem to "make sense" to us humans?

What difference does the non-linearity you use have on the solution your network arrives at?

Is the manifold hypothesis correct?

Why do deep networks generalize so well, even when they have a parameter count large enough to literally memorize their training set?

Not to mention how all of the hype in deep learning is basically around image related tasks. Which convolutional nets just happen to be very useful for. You can essentially throw any configuration of a conv net at any given computer vision task, and you will get something decent.

Right now deep learning is essentially a collection of very powerful heuristics. We have a very weak theoretical grasp of whats happening. If I could summarize, we can build neural systems which learn by using optimization. But this is really kind of a hack. We don't really know how to make neural systems learn.