/sci/ - Science & Math

>>10234289
Life is glorified optimization.

>>10234289

Congratulations, you are red pilled now. Welcome to "people who actually had contact with engineering" club. Take your mandatory beer, you will need it during weeping over yet another boring regression model.

inb4 (under)graduate faggots and wannabe academics rage

>>10234289
Childhood is idolizing the Unabomber. Adulthood is idolizing /Pol/ Pot. What the Unabomber did was send out a few bombs. He was a gnat trying to kill an elephant. What /Pol/ Pot did was empty out the (((cities))), force a return to a pre-industrial agricultural economy and kill off the oversocialized intellectuals. He did everything the Unabomber wanted to do in an entire country.

>>10234289
Am I the only one whose interest actually rose when I figured that out? It changed AI from some pie in the sky popsci crap into something tangible and implementable to me.

>>10234594
>He was a gnat trying to kill an elephant. What /Pol/ Pot did was empty out the (((cities))), force a return to a pre-industrial agricultural economy and kill off the oversocialized intellectuals.
You forgot "while keeping a modern standard of living for himself and arming his goons with modern weaponry".

Pol Pot was nothing but a tyrannical, childish megalomaniac, useful idiot chink plant who wouldn´t have been able to differentiate between purely egotistical actions and the actions of an ideologically devoted man.

>>10234818
Fuck off foureyes lmao
If you like school so much why not pay a visit to s-19 :^)))))

>>10234818
>arming his goons with modern weaponry
Ted does not consider the use of modern weaponry to overthrow the industrial system to be hypocritical. In fact, the use of modern weaponry is required to do so, no where does he say that people have to fight industrialists with sticks.

>>10234289
because if you look at the results in what they applied it to, and have enough imagination to envision how that transposed to other areas, you can see vast potential power. The problem is in reality it can't really transpose out of what turn out to be special case problems.
You have a gay status based mindset "it's important because it's hard and only the smartiest smartypants can understand it so i want to show off to myself by learning it". instead of "it's important because it can do a lot". A mature person *would* be excited if it turned out greatly complex reasoning intelligence could be reduced to simple optimisation.

>>10234818
CRINGE.

>>10234289

All math is just glorified counting.

>>10235117
except its literally non-convex retard

>>10234289
its lets you publish

>>10235124

A lot of time the cost functions can be modified to be convex.

But optimization is cool.

>>10234289
Because it solves some interesting problems

>>10235132
Maybe like 10 years ago, no one cares anymore

>>10234294
bruh

>>10234294
Prove it

>>10234321
Based. Its just so easy to make money with this kind of knowledge though. Everybody wants to slap the 'AI' sticker on their product so they need some basic models thrown in on it.

>It's glorified optimization
Did you seriously think we would develop any new techniques when we can simply rebrand old stuff?

>>10235446
the functions being optimized are new
improved optimization methods are still being developed

>>10234289
Why is ted so based

>>10234289
CS Terminology is always souped up with memewords targeted for brainlets
>Deep learning
>Cellular automatons
>Ubiquitous Computing
>Cybernetics

>>10235514
He was pretty much proto Elliot Rodger except he chose to become infamous REEing at society instead of chad and stacy.

>>10235538

Tbh he was right.

>>10235431
>Everybody wants to slap the 'AI' sticker on their product so they need some basic models thrown in on it.
is linear regression good enough?

>>10234289
Because it is extremely useful.

>>10235806
Train a single neuron network with a linear activation. It's almost as good as linear regression but you can call it AI for top marketability

>>10235809
for you

>>10235846

He's a big dataset.

>>10234294
READ The Industrial Society And Its Future

>>10235531
Dynamic programming is my favorite such term.

>>10234289
People are excited because it works. There's no need to reinvent the wheel when what they're already doing is working. If it's not broke, don't fix it.

>>10234289

ML is not about the algorithms but about the drive to put sensors EVERYWHERE an the internet of things.

Data about everything anywhere correlate d by anything else everywhere ever.

Stop thinking like a STEM and star thinking like a JEW.

>>10234289
There are two kinds of people in this world OP.

People that recognise and understand that every conveivable problem can be reformulated into an optimisation problem, and stupid people.

Stupid people latch on to things like ML because they see powerful results without knowing what underlies the whole theory, when this whol buzz dies down they will still not understand standard form notation.

>>10234294
>>10234321
>>10235124
>>10235137
>>10235431
>>10235446

People of the first kind

>>10234597
>>10235028
>>10235125
>>10235142
>>10235472 (very)
>>10235871

Stupid people.

>>10235925
>I hate and don't understand ML, also I don't like to make money so I'll just call everyone who finds it remotely interesting stupid

>>10235939
Nigger I have ML library contributions on my CV, ML is extremely easy to understand it's literally just a list of applications you can learn.

I rightly think people who learn ML online and don't understand the underlying beyond the superficial algebraic systems are stupid (which they are).

>>10234885
>Ted does not consider the use of modern weaponry to overthrow the industrial system to be hypocritical.
Pol Pot didn´t spend his last years overthrowing anything, you imbecile - he ruled as dictator over a people he subjugated by brute force. Uncle Ted would not take kindly to such acts.

>>10235974
Ted K wasn't really an anarchist, he just was anti-civ and anti-tech. I would assume a hierarchical tribal structure would be his preference. Ruling over others with brute force is the premise for all primitive societies. There is little if any observable egalitarianism in most HG societies, its a social science meme that's been critiqued strongly by phys anth researchers for decades.

>>10235964
>ML library contributions on my CV
post them
>I rightly think people who learn ML online and don't understand the underlying beyond the superficial algebraic systems are stupid (which they are)
then why did you reply to people who simply said they find ml interesting you retard

>>10235908
>People that recognise and understand that every conveivable problem can be reformulated into an optimisation problem, and stupid people.
The firstmentioned claim is patently false - I´m saddened to say it, but you belong to the dullard camp.

>>10235975
> I would assume a hierarchical tribal structure would be his preference. Ruling over others with brute force is the premise for all primitive societies.
There is nothing tribal or primitive about subjugating millions of people in a quasi-communist agrarian society.

>>10235979
name one

>>10235988
Most areas of mathematics and physics past the undergraduate level.

>>10234289
Haha I felt the exact same way when I first started learning it.

>>10236009
be more specific retard

You cant argue with the results.

By that logic, human brain is a glorified optimization. Perhaps it is, but it works.

>>10235981
nigger I don't care at all about you and your retard quibbles, fuck off and don't ever reply to me again

>>10234289

>>10236052
I don't suppose you'd point us to the Voynich Manuscript if a question about identifying a plant arose.

What about the AI from neuroscience. The simulations.

>>10235925
>get a (You) in both lists
really makes you think

>>10236053
>implying Russel&Norvig is indecipherable
lel brainlet

>>10235979
No it's not you brainlet.

>>10236009
>math
Lrn2AutomatedTheoremProving
>Physics
Lrn2PrincipleOfLeastAction

If you don't know how they are related to optimisation it's because you're an ML thoery skimming mathlet.

>>10234294
inb4 >intelligent design

>>10235117
as a physicist who got into a programming job with minimal programming experience, this was literally me when learning it.

>tfw compscis think potential wells are the height of sophistication

holy fuggn gek

>>10234289
Why do people keep posting things like this as if it's insightful. What did you expect it to be, magic?

>>10234289
>X is all the buzz
>Start learning X
Hah, it's glorified Y!
Like why the fuck is anybody excited about X when you can just do Y amirite

>>10236721
>automated theorem proving
what a grand way to expose your brainletism

>>10235151
>no one cares about convexity
fuck off brainlet

>>10237208
neural net losses are not even close to convex

>>10237491
No shit, brainlet. The nonconvexity complicates things quite a bit.

>>10237502

Not really, I think convexity is kind if irrelevant for ANN since ultimately what's convex for your training data doesn't make it work for all data.

Basically convexity is a meme.

>>10237838
>since ultimately what's convex for your training data doesn't make it work for all data.
What?
>Basically convexity is a meme.
A glimpse in to the mind of a "machine learning" student.

>Sex is all the buz
>start learning about it

It's glorified rubbing. Like why the fuck are people so exited about it?

>>10237884
Not that anon, but he's not wrong. You don't actually want to find the global minimum of the loss as it will probably not general well to new data. This is another reason people like small minibatch updates for the extra randomness.
Lots of random jumps vaguely in the direction of a better objective >> closely following a smooth gradient like a pussy

>>10237200
https://en.m.wikipedia.org/wiki/Automated_reasoning

Read up and shut up.

>>10237491
You do realize that 60% of the field of convex programming is about finding convex reformulations and approximations of non-convex programming problems right?

What am I saying, you're an ML mathlet and engilet, probavly did a brainlet physics degree so of course you don't even know the difference between standard and canonical forms yet.

>>10238512
Jesus christ, you brainlets are literally reliving early nature heursitic memes.

Go read optimization papers from 2 decades ago, you'll make so many "breakthroughs".

>>10234289
then write some optimization algorithms and BTFO every NN out there anon.

You can't? Too bad, no millions for you then. I guess machine learning isn't just glorified optimization after all...

>>10238710
They're already freely available in literature. NNfags are just too brainlet to use it.

>>10238794
>hurr it's all already there and I don't like to get rich
vs
>i'm pulling shit out of my ass and don't know anything about machine learning

>>10238797
There are other, better ways to get rich using the same algorithms you stupid ignorant little shit. Controller design in petroleum. Drug discovery in Pharma. Investment portfolio optimization etc etc.

ML is the only way you and the hordes of pajeets know about because you'v3 never had a proper education.

Most people in the field don't care money anyway, they have cosy academic jobs on top of their royalties from algo implementations and work on the most important problems in world.

Why would they care some stupid fucking meme mostly used by losers trying to drip pennies from consumer data?

>>10238704
>>10238709
you are usually optimizing a function you don't really care about, training loss. you don't actually want a global minimum of this function, this is why things like early stopping exist

>>10234294
based and dare I say it, redpilled
>>10234289
fuq did you expect moran

>>10238876
>I-I don't care about m-money!

>>10238709
nobody cares about the fact that some mathematician described the process of creating an AI in mathematical terms 50 years ago

People care about getting an AI that will actually take their selfie and draw dog appendages onto their face, for real.

You probably think building bridges is futile too since one can calculate a bridge on paper.

>>10239454
Most modern optimization algorithms don't always care solely about the global minimum.

They find all the mimima quickly, which is needed in applications like energy surface exploration. In addition to using this in various graph structures you have various tools like certificates of optimality or other tools to help you explore the functions surface.

>>10239492
I never said that. I said there are better ways to make money. ML is shit money, but I don't expect a pajeet who's only seen 80k to understand that a 120k salary is a pittance.

>>10239497
>"""""people"""""

Ftfy. They are not people that people like us care about. They are miserable creatures that get left behind on a starvation diet in a worn torn country while Operation Paperclip provides the rest of us with cosy salaried jobs and a new home in freedomland.

People with real money, power and intellect don't care about stupid shit like that.

>>10239675
>in a worn torn country
Love it.

>>10239675
>better ways to make money
like what

>>10239693
The guy who developed BARON makes millions every year on royalties and sales.

That's selling pure optimization algos, not even the vast number of applications yet.

Your problem in life is you're chasing buzz jobs for the salary not realizing how those vaguely defined bubbles will burst in 5 years anyway and then you'll need a new career in an oversaturates market.

Much like material scientists vs people who studied pure nanotechnology degrees who are all unemployed.

>>10239709
>pulling shit out of his ass
post one source that states he's a millionaire

>>10239675
>People with real money, power and intellect don't care about stupid shit like that.
That's besides the point. What I am saying is that having a highly general mathematical concept for something is not the same as having a working technology. AI might not be anything theoretically magic or shocking (whoah who woulda thunk), but it still takes work and ingenuity to create practical implementations for it under real world constraints, which is ultimately the only thing that matters. The end user only cares about the product.

There is a reason why so many mathematicians are commies even though communism will never work. They treat "real world constraints" as something disgusting, preferring limitless generality.

>>10234321
i went to a conference w a lot of neural net ppl there, even people from google. If you pushed them in colloquium they would fold in an instant and acknowledge this, but then immediately go back to touting their second coming of AI during their talk. I guess since they rake in the money, its a reasonable science career, so long as you are ok w the fact that you are spending your entire life contributing absolutely nothing to human understanding and doing something which is completely redundant.

>>10235531
the only thing I would take off your list is cybernetics, which if studied as wiener intended, would actually be a rich subject. except most cs ppl are retards and could probably not follow the computations involved in even the early basic demonstrations of it by wiener.

>>10234294
we live.......in a society

Because cheap hardware and HUGE datasets makes previously unfeasible ML algorithms, feasible.