Is machine learning a meme?
I'm in grad school studying ML with the belief that it will be a dominating field in the near future.
Is machine learning just another string theory in that it's all hype and too many people study it, or is it actually the revolution it's promising? Will there be as many unemployed machine learning experts as there are string theorists?
>>7750541
It's basically a shitty empirical model for lazy people with a ton of resources at their disposal.
10 years ago, AI was all the hype. Now it's ML. Soon it'll be a new buzzword.
>>7750541
Everything is a meme
>>7750541
>I'm in grad school studying ML
Somehow I find that hard to believe given this thread.
The glorious revolution of GNR.
Genetics
Nanotechnology
Robotics
With above human intelligence AI, the AI will improve upon itself and acheive exponential gains with each improvement. The AI will improve upon itself faster and faster because of the better tools at it's disposal. We will become slaves to this new AI. The AI will order trillions of nanobots to acheive domination of the earth and eventually the whole universe.
>>7750541
There is a lot to learn about ANN, I expect the field to be around for a long time. It already has practical applications but we still haven't really explained how it works and behaves. So unlike string theory in which no practical application will be made possible any time soon it already has practical application which make it financially more feasible to grow. So going into machine learning should give a far bigger change at landing a job then some self proclaimed Motl or Sheldon ever will with there hypothetical yau calabi manifolds.
>>7750546
This is kind of true. ML revolves around abusing computer resources to train an excessively general model to accomplish a broad variety of tasks.
This is not entirely a bad thing, it means that you can save time by writing one program that is capable of many applications. But in almost all cases ML is saving human time, and not accomplishing anything as well (in numerical accuracy) or as quickly (in computational time) as writing a problem-specific approach.
>>7750541
>Is machine learning a meme?
Yes
>I'm in grad school studying ML with the belief that it will be a dominating field in the near future.
I'm so sorry.
>>7750541
Machine Learning is a great engineering tool, a lot of stuff can be done with ML that we can't duplicate or understand. It mostly empirical and doesn't require much knowledge in the field you want to apply it to. All you need is enough data.
So you will get probably a good job in the industry with a PhD in ML (if you aren't an idiot), but ML is a horrible tool to understand a subject. Just look at all the NLP work, we don't know what the language model is that humans use, but can do okish automatic translation. ML can approximate a lot, but it won't explain shit.
Also some people suggest that in the main fields where ML is used (CV, NLP), we are near to the point where the "throw more data at it" approach doesn't give better performance.
>>7750541
Yes, it's a meme, but not just a bullshit 4chan meme, it's a business meme as well, if you specialize in machine learning you will be able to find plenty of jobs working for people who don't quite get what it is but think it's important.
>>7750610
You say that like there are machine learning jobs out there. In fact there are almost none, it's just a buzzword that gets thrown around.
The work is in statistical computing in general.
>>7750553
This is the correct answer.
Even the way we express most things in math is technically a meme.
>spend lots of years on string theory and get nowhere
>now sitting at -$200k
>spend lots of years on ML research/applications and get nowhere
>work as a quant sitting at +$200k
Are all these data scientist things just a maymay? Apart from implementing regression and other Statistics 101 concepts, what the hell can you even do (can someone explain in simple terms)?
>>7750690
You won't get hired as a quant by studying ML. Almost nobody wants ML quants. Again, you just want general statistical computing skills.
Also getting hired as a quant is really fucking hard. People who wave it away as "oh you have a math degree so wall street will eat you up" haven't looked at the market since the 90s and are just passing down knowledge. The quant job market is fucking brutal.
>>7750541
People are generally shit at applying the technology, but a good model can make mad dosh.
What is the next meme field?
If string theory was a meme field and ML is the current meme field, what do you guys foresee being the meme field in 5 years?
>>7751330
curious about this as well, what is the next meme subfield, /sci/?
>>7750610
>Yes, it's a meme, but not just a bullshit 4chan meme, it's a business meme as well
A lot like Java.
>>7750541
What will the machine need someone like you for once google releases their beast?
>>7750546
Yeah. But our shitty empirical models are outperforming and invalidating decades worth of computer vision research.
>>7751734
Nah man. Java is the shit. Some pretty sophisticated OOP architectures can be handled very eloquently using that language.
>>7750541
ML is the only reasonable application of NNs to practical problems. It is easy enough that a protocol can be designed and refined by a single person. It is a meme in that it is not artificial intelligence in the general sense, and that it needs to be trained in order to be rendered applicable.
Its biggest 'memetic' is that it is an 'approximation' despite having been proven to outperform a human after sufficient refinement of the learning protocols.
>>7750541
Everyone in this thread is retarded and not in grad school. You can make 100k+ a year easily if you sell out and program stock trading algorithms. Also from a science perspective, nearly every tool we have for interpreting data involves machine learning.
There are a lot of methods across engineering that are basically the same problem stated in different ways. Machine Learning, Controls/Optimization, System Identification and Filtering are the same thing. If one is a meme, they are all a meme.
>>7751925
>OOP
>architectures
top kek
>>7751943
Dude. Don't tell me you haven't seen some incredibly impressive shit pulled in java.
>>7751978
I have seen shit pulled in Java. Lots of shit. Like everything coded in Java.
>>7751978
If by impressive you mean in the sense that I can't believe anyone would go through that, yes. I once saw a guy who programmed a game in Java. His wife even divorced him over it. His name was Notch. The game is exactly what you would expect from a java game. Shit graphics, slow as shit, uses obscene amounts of ram, buggy as shit.
>>7750541
Is a masters enough to meme people in machine learning? I don't know if I'm autistic enough to do a PhD and I just want dosh.
>>7752043
of course it's enough
PhD is just for people who have nothing to do with their lives
>>7751994
you sound overwhelmed by the power and superiority of java. You should seek psychiatric assistance anon.
>>7750541
Machine learning is just another buzzword for statistics. There's also "business intelligence" which again is statistics. "Statistical learning" itself is just curve fitting which is just numerical analysis. Basically all these new memes end up being numerical analysis.
>>7750708
>The quant job market is fucking brutal.
What ? They even hire poor-level grad students with a limited knowledge of stochastics.
It might not be the case in the US but here in Europe (at least FR/UK) it's clearly opened and hiring
>>7751728
Memeology
>>7750572
this is called the grey goo death
>>7752521
numerical analysis is itself the old meme
HOTT will be the new meme .
>>7752521
statistics is fitting existant models to data
machine learning is automatically creating models that fit data
see the subtelty there?
>>7752523
encore un désillusionné qui croit qu'il va finir riche en allant à la city alors que c'est plus bouché que l'anus de sasha grey pendant un gangbang
trouve un vrai job
>>7750553
Literally this
