Is machine learning fucking overhyped among normies?

I don't know, I think using ML for AIs would be fun.

>>62404002
That's how all of marketing works and always has.
You intentionally pick a clever sounding word most people don't know the meaning of. Because then they can use their half-(or rather less)-knowledge about it and supplement it with what they'd like it to be to arrive at a conclusion that leads to a sale.

File: NUP_169190_1090.jpg (1MB, 3600x2396px) Image search: [Google]

1MB, 3600x2396px

>>62404002
Yeah machine learning and neural networks are the new "omg nerds and kool technology" terms.

Every pajeet using them think they're the smartest fags on the planet.

I can totally see a new Mr Robot season where Elliot autistically monologues quoting machine learning and everyone looks at him thinking "oooh how smart"

>>62404039
Remember "Blast Processing" from SEGA anyone?

People tend to get confused when you tell them that "ML" is basically just automated statistics.

>>62404002
>>62404039
I don't know honestly. Yeah, if Apple does it it will work for sure, but as OP said, will anyone really buy a game that """uses machine learning""" if it sucks ass?

>>62404036
It would be one hell of a work to make it decent though. An AI learning from the player could either behave in ways the pajeet who developed the game can't predict causing weird behavior, or become too quickly too strong so that most players don't have a chance.

>>62404075
This! As a statistics PhD I always call BS when I see all this retarded hype.

It's simply using what we've been doing since always but with algorithms processing the results.

>>62404087
>will anyone really buy a game that """uses machine learning""" if it sucks ass?

Probably not. But it's almost always one of the big players doing something "new" and then everyone else trying to ride their coattail with varying degrees of success.

I hope CRISPR will be the next big pop-sci trend and idiots will stop talking about AI.

>>62404099
That's it, if the new Halo does it with decent results it will probably work (I said Halo cause they actually used it in development but never used it as a marketing word), but what about some crappy indie game developed by sub-pajeets?

>>62404002
>iphone machine learns to adapt to your face over time
>learns wrong and stops unlocking your phone
or
>facial injury, stops unlocking your phone

>>62404114
>but what about some crappy indie game developed by sub-pajeets?

They'll still get some sales from gullible idiots who would have otherwise not bought their game.
Vidya games have a large branch that does nothing but skim off easy profits by aping more successful franchises. Just look at all the minecraft clones. None of them led to their developer becoming as rich as Notch, but in a good amount of cases they let to a couple thousand for minor effort.

>>62404129
they'll probably just still have the regular key codes on top of it, right?

>>62404136
they god rid of the headphone jack

just think about that anon

>>62404002
It's probably the new buzzword like "big data" and "cloud"

>>62404168
This!!!

>>62404129
>shave
>phone stops unlocking

>>62404129
Yeah, people doesn't realize ML is unreliable at best. Heck /b/tards managed to make Microsoft's AI a Nazi, just think about it.

>>62404134
I sometimes wonder: do retards who buy that crap actually exist? I mean, I think it only has a market on mobile stores where people looking for the original game may be fooled by clones, but being that Minecraft is there as well, who would be so retard to buy the clones instead?

>>62404186
>dont shave
>phone stops unlocking

>>62404186
>Get bruised while shaving or for some other reason
>Phone stops unlocking

>>62404195
>who would be so retard to buy the clones instead?
people who have spent hundreds of hours on Minecraft already and think they can relive their first hours playing it again if they pick up one of those clones.
for other games where they copy a more expensive game it's to grab poorfag countries as audience who cannot afford the original

>>62404114
Sorry if I'm a newbie, but what does sub-pajeet mean? I only know pajeet means an Indian with bad hygiene who poos in loo (according to Urban dictionary)

>>62404227
Pajeet is actually used to mock bad indian programmers, and the meme is generally that they DON'T poo in the loo.

Sub-pajeet means koders who are even worse than pajeets, often residuals of shitty programming schools who are too retarded to go to a decent college or to even be self-taught. An incredibly high percentage of these are in the indie gaming field, thus the proliferation of scam game development schools and shitty buggy clone games.

>>62404197
>>62404186
Your best bet is to do the face scan when you're fully shaved, and then remember to shave every day lol

>Inb4 you forget a singe hair under your nose in your whole face and it doesn't unlock

>>62404195
>Yeah, people doesn't realize ML is unreliable at best.
I don't think that's really the issue, and in fact I am probably wrong in what I said.

What it most likely means is that when it is saving/checking 30,000 points on your face it's allowing a reasonable margin of error, so when get over that margin of error and it unlocks it will save that error margin and use it to adjust the saved data so that as your face may change slowly over time the saved pattern adjusts through the error margins of successful unlocks.

And that's probably fine. It is an application of machine learning. It's not magic. It's not particularly special. But again they show someone basically joining ISIS and apparently it still recognizing them, which is quite a bold claim, covering up that much of the face and still recognizing him.

In one of the demo videos the presenter also mentioned the possibility of being able to subvert the unlock with a picture or something and not being sure if Apple has done anything to prevent that (this was an Apple employee), and that was an alarm bell for me.
They should be sure. If they're taking away fingerprint completely (which wasn't that secure either, really) and replacing it with something that MAY still be able to be subverted with just a picture of you then they're completely out of their minds.

>>62404186
>>62404197
>>62404212
>A lookalike steals your phone
>It unlocks

>>62404268
>Your best bet is to do the face scan when you're fully shaved, and then remember to shave every day lol
Apple, we're totally not deciding how people should live.

>>62404002

Short answer is yes.

But I do think ML will be really nice for incident resolution for end users, NOC and SOCs alike.

File: obama-lookalike-Ilham-Anas.jpg (100KB, 640x417px) Image search: [Google]

100KB, 640x417px

>>62404287
>>62404303
So this guy can unlock Obama's phone

It's quite frightening that the Apple employee didn't know if Apple made something to avoid unlocking with a photograph...

>>62404324
Is a 3D scan of the face, so a photo can't work, but maybe a 3D print of a face?

>>62404333
With 3D printers getting more and more affordable, it's something to consider...

>>62404333
3D scan is when it memorizes it, I don't know if it performs one on each unlock. Wouldn't it be quite resource intensive?

>>62404324
Maybe. Maybe not if it takes his ears into account.

The fact is that it doesn't look at a face the same way we look at a face.

It's like saying it wouldn't work with Asians because they all look alike. They don't, but that's how they appear to some people.

>>62404287
>In one of the demo videos the presenter also mentioned the possibility of being able to subvert the unlock with a picture or something and not being sure if Apple has done anything to prevent that (this was an Apple employee), and that was an alarm bell for me.

Didn't they claim it's paying attention to the face being 3D to prevent that?

is all that effort really worth it just to avoid typing in a couple numbers?

>>62404362
Well apparently, unless is all marketing buzz they have this Bionic processor with a neural engine ( I don't know what they mean by that), precisely to handle what they call intense computing load of Face ID.

>>62404367
What is 3D to a single lens camera?

>>62404380
Is all that effort really worth it just to avoid pressing the home button with integrated fingerprint scanner?

>>62404002
It's overhyped in general.

>>62404002
>Most retarded ones being those who want to use ML in indie videogames

For what?

t. Data """scientist""" who use ML

>>62404404
not him, but iirc I read some stuff about them wanting to use ML to make the environment more dynamic and capable of doing things it wasn't explicitly told to do

which sounds like a recipe for disaster

>>62404385
The way they used the "neural engine" buzzword was the entire topic of the thead. Why would users give a fucking shit if it uses neural networks, machine learning, or dog testicles, the important is that it's reliable and safe in whatever it does.

Now I could trust this since it's Apple and while never being an Apple fan (I never had a Mac or iPhone but my family members and friends did so I'm familiar) I know they have an history of great hardware and software, but most people who claim to "work with ML" don't really know a thing of what they're doing. They just call APIs and then lick their shit-encrusted mustache in satisfaction.

>>62404422
ML doesn't work like that.

>>62404428
when has that ever stopped marketing?

>>62404422
It doesn't work like that, saying "making programs do things they weren't programmed for" is marketing BS. Computers ONLY do what they're proframmed for, ML is just giving them algorithms to adapt to new data they're given.

>>62404305
If this is gonna stop le beards hipster trend, I'm gonna shill it.

>>62404445
and this whole thread is about marketing bullshit, isn't it?

>>62404251
Oh I see, I learned something.
I guess I could actually know some of them irl... Mostly people who dropped out of college for laziness/retardedness.

File: fe974dee-ee4a-46b9-a1f5-5787ce0881f9..png (7KB, 160x160px) Image search: [Google]

7KB, 160x160px

Is machine learning even real ? I thought it was a scifi joke/meme, that obviously doesn't exists, like space elevators or cyborgs. How can a computer even learn ? Am I falling for a bait or something ?

>>62404462
Yeah just answering the other anon now, since all this is marketing BS it's better to let him know

>>62404490
It basically isn't real in the way they depict it. I'm no expert but in short:

You just give a program some instructions about how to react to the data he's given and adapt his behavior consequently. But this is never 100% reliable without a constant supervision that tells him if he's going in the wrong direction.

>>62404490
I would like to have some good resources thst explain what ML really is, since there's all this marketing disinformation

>>62404490
Is learning even real?

Your question is wrong.

>>62404533
https://en.wikipedia.org/wiki/Machine_learning

knock yourself out

>>62404490
just read the wikipedia, it explains it pretty well.

https://en.wikipedia.org/wiki/Machine_learning

>>62404552
>>62404576
I just don't like how the Wikipedia page too uses too much "making machines think™" rethorics instead of going straight to the point of what ML ACTUALLY does.

>>62404096
Lul, this is some solid coping.

Linear regression is to a CNN as logic gates are to an ALU

It's braindead shit when you look at the minutiae but being able to set up the system so it works well end to end is where it gets dank.

And getting into RL, where youre automating the experiment generation as well is where it gets even more fucked.

Manual Stats is like zeroth order inference (fit my model bitch), CNNs are like first order(I don't know the model so fit whatever you can), and RL is like second order (I don't know shit, here's a black box with some buttons on it, come back when you win a Go tournament)

>>62404456
It won't.
It'll just encourage people to keep the shitty facial hair styles they have.
>i need it to unlock my phone!

>>62404520
Kinda there but not quite right.

You can't just take data and pipe it into a machine learning system and expect the right output because different types of data have different ways they are distributed.

However, if you have similar sets of data then you can run the same type of machine learning algorithm on them, so you don't have to make sure it is actually working correctly and producing useful output.

>>62404002
I think though instead of "virtual neurons" using random numbers if it could run through pre-built data to return a solid 0 or 1 it might actually be something real

File: 1502438575766.jpg (28KB, 512x223px) Image search: [Google]

28KB, 512x223px

Tell me why I should listen to the contrarians on /g/ instead of a self made billionaire that created an AI to play the most complicated video game currently out better than any human being on the planet?

>>62404742
Billionaires do not make money being innocent

>>62404422
>I read some stuff about them wanting to use ML to make the environment more dynamic and capable of doing things it wasn't explicitly told to do

Based on what? ML are algorithms to either predict values (different kinds of regression algorithms) or to classify stuff.

Are they going to get some data in and then send it through some ML algorithm and then generate the environment based on that? Sounds like an extra unneccesary step. I don't see the value of predicting anything or classifying anything in that case.

>>62404742
The dota2 bot is actual fake news. It's a super contrived scenario where whoever has the best reaction speed will win. No planning, pure reaction, of course the computer will fucking win

>>62404773
Yeah it's fucking unnecessary in games. Roguelikes were made well before this jackshit.

>>62404776
This. That's why it's stupid to think about "powerful AIs" as something revolutionary in most games. AI can be made strong as hell just by giving her the actual data the gaym already has and lowering it's reaction time.

Elon Musk is an idiot.

>>62404445
>>62404422
>>62404428
>>62404789


Wait a minute (I am this guy >>62404773
& >>62404404).... I just thought of something, what about making a game with some sort of reinforcement learning algorithm that makes the game difficult for the player, but not TOO difficult. Just the sweet spot. So that it does not matter how skilled people are they will still experience the game as difficult in about the same amount. Or make it slightly easier when he dies alot and so on. THAT would be a useful use of ML in games.

>>62404776
Then is not fake news, it was blown out of proportions, exaggerated, but the thing exists, it works in its scenario, so is not fake, how is it fake?

>>62404808
Could just have an easy mode and hard mode, when the AI score is too high it can switch to easy mode, to account for human fatigue perhaps

Machine learning do some quite impressive shit and do things you don't normally expect a computer to do well at all, but seems like the general public is not very aware of what it can or can't do.

>>62404832
It's fake news because they weren't playing dota. Dota is a team game, not a one on one skirmish.

>>62404808
Still kinda overkill. It just needs a set of difficulty modes and to automatically switch between them when the players wins or loses too much.

>>62404096
Some grad books about stats.

>>62404870
It would be cool to make a neural network from scratch with code without flipping out at all the calculus the code typically involves at this time

>>62404868
Not even sub-pajeet indie (more like Indian) game developers. There was these guys I met at a developer conference who don't even know C (they use C++ as a clunkier Java basically), never actually had a job or released a game (only one of them worked as a Java code monkey), have 0 CS culture, but claim they're working on a "Machine learning based game". From some screens it seems they're just using some framework APIs for God knows what jackshit.

>>62404688
>Linear regression is to a CNN as logic gates are to an ALU
you jest rekt yourself fampai,
as floating point math takes 2-2.5 order of magnitude more time, if you were to do it on ALU

>>62404924
http://sgsa.berkeley.edu/current-students/recommended-books
There's some nice ones here. And funnily they also have a ML category, further proving our point.

>>62404075
>People
Even CS students get triggered by this.

>>62404947
This

>>62404927
If CS students do, imagine sub-pajeets.

File: 44229962_l-Copy.jpg (108KB, 847x565px) Image search: [Google]

108KB, 847x565px

>>62404689
>So ironic phone doesn't recognize me without le manly beard I guess I'm just made to be with my super ironic facial hair xD

>>62405011
That pic is basically everything wrong with 2017

>>62404941
Thanks
Foundations of Modern Probability by Olav Kallenberg
Looks good.

>>62404002
Well isnt it true that ml is getting bigger and bigger and is the closest thing to ai that we have? Thats the definition of intelligence, being able to acquire and apply skills. Ml is still limited but if we ever want to have ai then ml is an essential component

>>62405066
I've studied a lot on that, it's really a good one!

But if you wanted something even more for beginners, to get your feet wet and have a general introduction while touching intermediate level concepts, this is good too:
https://www.amazon.com/Introductory-Statistics-Third-Sheldon-Ross/dp/0123743885

>>62405149
We just don't want people to believe it's magic and dismiss it for what it really is. It's not even intelligence, it can be reliable only if limited to a really specific area (like Apple's face recognition)

>>62404043
How did someone who looks so fucking normal and unimpressive get hired and get famous?

>>62404002
Fucking marketers and their fucking buzzwords.

File: 2016-06-27_07.jpg (99KB, 1279x2275px) Image search: [Google]

99KB, 1279x2275px

I love to see them try and fail in fields where human intellect is the deciding factor in the outcome.

They unironically think they can algorithmize languages and can make a machine that translates between them well. In the meanwhile google googles, the PR app for machine learning at google doesn't recognize Nail Scissors and visually similar images return a spoon. This is just fucking embarrassing.

>>62404533
All machine learning pretty much comes down to statistics.

It is basically applied stats. It goes like this:

Overweight Male DIED OF HEART ATTACK
Overweight Male DIED OF HEART ATTACK
Underweight Male DIED OF CANCER
Overweight Male DIED OF CANCER
Underweight Male DIED OF CANCER
Overweight Male DIED OF HEART ATTACK

So if this is the data set, we can find the probabilities of a lot of things.

Probability you will die of a heart attack, given that you are overweight, from this dataset, it is 75%. (You can see why huge datasets are needed to train machine learning algorithms.)

You can see the downfall as well from small datasets, P(die of cancer | underweight) = 1

You can obviously get way more complex, but it pretty much all comes down to this sort of thinking.

I wonder how Apple created a "neural engine". And whether it is just like the taptic engine, an improved vibrator

>>62405244
Look at scenes of the show. He looks super autistic and it fits the character. He's basically a stereotypical sociopath computer genius.

Also, he's indian.

>>62405298
The algorithm has no opinions, no preset beliefs. Once a training set is processed it does not remember what it learned in the next training set. Just a glorified rat brain

>>62405244
Some girls say he's hot

Marketing BS for wannabes

File: rioterselfie.jpg (70KB, 800x450px) Image search: [Google]

70KB, 800x450px

>>62405049

>>62405341
>it does not remember what it learned in the next training set.
What are kernels.

>>62404947

No wonder, CS students are brainlets.

>>62404742
I know you are just baiting but I want to reply anyway.

Appeal to authority fallacy.

He did not even create the algorithm which is not an AI. And when an algorithm can beat a human, it is a statement about the game's simplicity, not humans. MOBAtards got btfo which is great.

Elon Musk has many stupid beliefs. I have great respect for him and his achievements but that does not make him infallible. While only he himself knows for certain, I certainly think it's possible that him partaking in the climate change hysteria and proposing we colonize mars in the next couple years is merely a PR stunt. Tesla so far has only made huge deficits and depends on government aid to survive. This aid's continued supply is based on climate hysteria. SpaceX too is completely dependent on government help and it uses nonsensical goals like sending people to mars for publicity and marketing

>>62405305
I believe begins math matrix multiplication special chip, but couldn't found more information.

>>62405335
He's Egyptian

>>62404390
Probably has some laser or what not to get positional data. Like the kinect or similar devices

You'll never get a job doing ML and if you do, your employer will expect positive results from your research while in reality you'll end up spending a week running training your algorithm and ending up with overfitted data because of your shitty training data. You can never really debug the results because it's essentially black box testing.

>>62405283
How can it. A computer doesnt "see/ percieve" it only has an array of data that an algorythm tries to make sense off. Its like you would have to find something in 1mb of byte data.
Interfacing between the binary and the real world is very limited.

>>62405958
>You'll never get a job doing ML
funny because I have a friend who currently studies CS. he took a year break to work in ML related stuff and every company he applied to accepted him and offered good wage too

>>62405967
add a camera to the computer

your brain is literally the same. it does not see. it gets electrochemical signals from the eyes

>>62406066
Google algorithm found wood table plus object metal begins huge advance over last decade but Computer Vision still need a lot work.

>>62406088
yes. my point was the problem is not the lack of visual perception but bad data processing. surprise surprise, turns out you can't replicate the human brain with a couple lines of code, some statistical data and a bunch of transistors

>>62404002

Machine learning = automated statistics

The CLOUD = someone elses computer


The tech industry loves to spawn these meme named so that everything sounds cooler than it really is.

Example can you imagine if "cloud" providers would have started marketing their services with name "someone elses computer" instead with name "cloud"?

Even normies would stop using cloud storage after that.

>>62404002
"Machine learning" and "neural network" are buzzwords. They only exist because hack journalists needed a name that sounded sexier than what the process actually is, i.e. linear optimisation.

File: quote-if-the-brain-were-so-simple-we-could-understand-it-we-would-be-so-simple-we-couldn-t-lyall-watson-30-83-03.jpg (54KB, 850x400px) Image search: [Google]

54KB, 850x400px

>>62406066
Those electrochemical signals aren't binary, and the physical neuron-based structure of your brain can't be replicated by silicon, even if we did understand how it worked (which we don't).

>>62406258
>if computers were so simple that other computers could simulate them, those computers would be so simple they couldn't.
that's how he sounds to be honest

>>62406313
But he's still right. A computer can only be perfectly simulated by a more powerful computer. The analogous situation would be some hyper-intelligent biological alien which can fully understand human brains. Which does nothing to change the fact that humans still can't fully understand themselves (nor could the alien fully understand itself).

>>62406258
>Those electrochemical signals aren't binary
I'm not trying to be pedantic for any reason other than curiosity, but does this matter? At the end of the day something is either present or not, be it an electrical signal or some kind of chemical, even if there is some variance in volume this can always be abstracted in implementation then replicated/represented again.

I think the problem is less "how do we represent this" and more "what do we represent".