What does Veeky Forums think of Artificial Intelligence/Machine Learning? Is it CS tier or better?

Most AI/ML people are literally people who have studied CS.

I mean, I don't think it's a meme. It's here to stay. If it interests you, you should learn a bit online. Then decide if you want to continue with ML classes / MS degree / job in ML.

Is it CS tier? Who cares. You'll have a job doing something interesting.

I don't care whether it's CS-tier or not, but I wanted to know how much shitposting it attracts compared to CS.

literally me holy shit and that's years before amadeus even came

Actually this board could use some more ML memes

You've got other areas of math in there as well, such as convex analysis.

Well, if you follow the derivations, you'll probably see that they're trying to find out ways they can predict the output of a black box.
The boring parts of ML are when they try to get 0.1% more accuracy on classifying the MNIST dataset.

Most good AI/ML people are literally people who have studied maths/physics.

Geoff Hinton and Yoshua Bengio are doing some fucking groundbreaking shit. Truly cutting edge.

Everyone else is just training neural nets on the latest datasets and making tweaks here and there.

Why are math/physics people so insecure?

Your constant need for approval makes me believe that your degree must truly be useless when it comes to finding work.

You're a fucking retard.

AI is not the same as ML as many people here seem to believe. ML is the modern approach to many AI problems involving "learning".

AI involves searching state spaces, logic (automated theorem proving, planning, etc), and of course ML. There are many approaches to AI, and ML isn't the end all solution by any means.

Also AI is a branch of computer science. ML can be seen as a branch of computer science which borrows ideas from probability, optimization, statistics.

Most AI/ML courses are offered by the CS department, but in some schools there are ML departments (I believe CMU has one).

Found the CS undergrads

>implying that's an insult
How does it feel that on average CS majors make more money than physics/math majors? Veeky Forums memes don't apply in the real world. Good luck finding a decent job with a physics/math degree kek

My interest in it was raised since an official from the US government came to one of my classes for a lecture on inter/national security and said that they've been making significant investments in machine learning for adaptive anti-hacking as a way to raise their shields on the new front that is cyber security.

An adaptive antivirus sounds like some cool shit.

>I can't tell the difference from degree titles and job titles

kek

Considering I'm a physics grad doing a PhD in machine learning (which I got into because of my experience in my first graduate job)... I feel pretty fucking good, user.
I feel good knowing I'll make more than any CS grad, and I feel good knowing my department were clamoring to get me to undertake a PhD, and I feel good knowing I'm not gonna waste my life as a front-end developer with no potential for career advancement.

>I would add that ML has only recently become popular because of the recent abundance of data and increased computing power.
So much this. In the 90s NNs were dead because it is computationally awful and you couldn't use big ones, then in the 2010s they discovered that if you have googles server farm than you don't give a fuck about that and called it "deep learning" aka "big neural networks"

>It makes me feel like they are all just applying NN's to new datasets and praying nobody notices they didn't do anything special while they slip out with a PhD.
Some profs in our department even admitted that a CV or NLP PhD is only about finding new features you can use the same stale ML methods on. Sure, finding features is hard and requires a lot of domain knowledge, but it's still a joke.

I feel there's a small distinction to be made between ANN and DNN... With DNN you really aren't gonna have much luck predicting the outcome, and you have more hyperparameters to worry about, but the results are much better.

Is going into Machine Learning a viable option after getting a degree in mathematics?

It's pretty much one of the best options.

In what way, money or interesting research?

>solving intelligence is the one problem most worth working on
>solve intelligence = solve everything != meme
>contributing even a little to the field is life well lived

Yeah AI/ML is better than CS

Well, with a maths background you're better suited to machine learning than almost any other graduate. The money is also pretty fantastic, but this is likely due to the data science bubble.
Most companies are trying to hire data analysts (basic reports/presentations, summary statistics, and graphs) and don't realise it, while at the same time many data analysts are billing themselves as data scientists for extra pay.
A data scientist is sort of a cheap actuarial scientist or quantitative analyst that can code. The reason why they're 'cheaper' is because good grades and extra qualifications are required to be a quant or actuary, but not a data scientist.
I think the key distinction between analysts/quants/actuaries and data scientists is the emphasis on machine learning. Machine learning is basically statistics at scale, and so to many businesses there is little distinction between someone who can do statistics and someone who can do machine learning.
Outside of business, you will make less money, but the research is significantly more interesting (i.e. you're not just cleaning data and presenting summary statistics). Personally, I think and are right in that actually studying the algorithms is pretty dull research, but the applications of machine learning are really the most interesting part, e.g. discovering robust biomarkers for illnesses, creating self-driving cars, fraud/cancer/whatever detection, weather/climate modelling, astrophysics, risk prediction, etc...
This kind of research requires domain knowledge from outside mathematics and machine learning, but learning that is part of the fun. It's likely that when the academic researc is solid enough, many of these applications of machine learning will become commercialised, and business will start to take over research in these areas (we're already seeing this with the big tech companies regarding self-driving cars and facial recognition), so ML jobs won't disappear any time soon.

this

every branch of science benefits from automated intelligence. It's amazing how many smart people don't realize this.

AI haters/doubters are usually just people who don't realize they are dualists.

>Some profs in our department even admitted that a CV or NLP PhD is only about finding new features you can use the same stale ML methods on. Sure, finding features is hard and requires a lot of domain knowledge, but it's still a joke.

Which is why Deep Learning is changing the landscape entirely in CV and NLP. Hand engineering features is dumb, let your model learn the features on its own. That's deep learning in a nutshell.

Awesome write up, thanks! In the long run do you think it will be as viable a career as a quants? I dont want to find out that in ten years the bubble pops and Ill get fucked. Do you have any recommendations of courses I can pick up in uni that will increase my chances of employability after graduating and in the worst case scenario of a popped bubble?

What's the best degree to get in order to get into ML? Math, Computer Science or something else? I don't have the best idea of what is best for ML.

Applied Mathematics is the best degree for machine learning by far.

Thanks user

>"Neural Networks" = Trying to make a helicopter, starting with a microwave oven, and tweaking it dramatically until it kinda works... badly, but still can't fly.
>"Deep Belief Nets and Deep Learning" = Trying to make a helicopter, starting with a vacuum cleaner, and having to tweak it marginally less than before, having waited for your tools to get slightly better, then concluding a vacuum cleaner is a better starting point than a microwave oven when building helicopters.

Welcome to the field of Machine Intelligence in 2017...

you are really bad at analogies...

Honestly? It doesn't matter, you would be good with a Math/CS/Physics/whatever as long as you take the useful courses and do something relevant to the subject in your MSc/PhD. Most of the things you will learn for ML aren't in undergrad anyway.

I think the 'data science' bubble is about businesses hearing that they need to hire data scientists but not being sure why, or what to do with them - so they just treat them as analysts. As I said, a lot of people call themselves data scientists because no one knows what they are (some obscure combination of business knowledge, machine learning, and programming), and very few people live up to the hype. Once businesses start to realise most data scientists won't return on investment, the demand will drop.
A person who knows stats is useful anywhere, and a person who knows programming will have no problem finding a job, and this very likely won't change, so try pick up some stats or CS modules if you're concerned about employability.

Thanks user, that is really helpful

He's as bad at analogies as I am at analogies

What kind of fields can't be automated easily?

Do we have a definite answer whether or not everything can be automated? Like a Godel's Incompleteness Theorem, but for ML?

>What kind of fields can't be automated easily?
Politician

>Politician

We could automate it, at least partially.

I am thinking of a system that has access to all the banks databases and counts all the money everyone has. The moment there is too much income inequality, we allow it to pass laws that raise taxes on the rich and lower them on the poor to balance things out.

When these laws are passed everyone who is involved in enforcing them gets an email.

Then when the balance has been found then it removes those laws to give the economy a fresh start. If we fuck up again then it will again put the laws in place.

The same could be done with a computer that measures pollution. All we have to do is allow it to measure something and then pre-write laws that would become actual law should be computer ever measure too much pollution, or income inequality, etc.

You understand that enforcing equality of outcome is an ideological thing, not political, right?

>Police
>Military
>Artistry
>Science
>Law
>Marketing
Just off the top of my head.

Machines can't learn, learning is a biological process.

Machines can learn, although it might be better to dub it "New Information Integration" rather than learning. Computers are introduced to a new variable/idea and their coding tells them to integrate it into a new database, or even alter their programming with respect to the new information they have just taken in.

Sounds legit, thanks again.

And how do you think your brain is any diffrent ?

You can emulate any biological process once it's understood

tell me about "intelligent systems"

is it a meme or not

...

You won't make more money to start, almost all research labs doesn't pay a lot.

You will if you start your own company or get hired as a senior ML researcher for a gigantic corp like say, Google but will take about 5-10yrs to get there even w/PhD. Google at first would pay you probably the bay area standard of 150k which is what a front end developer makes too. After your first productivity yearly review you undoubtedly will fail, as they seem to want to push out anybody to keep salaries low, and you will want out of Google but that's when you get the real money jumping into a startup.

There's also financial programming if you know stuff like martingale models/probability. That's where the huge, huge money is in Computer Science working the trading room floor putting together a trading strategy on the fly with C++ (these days they're using Haskell) and receiving your bonus of 2 million dollars a year.

You don't know what you're taking about.

>You won't make more money to start, almost all research labs doesn't pay a lot.
Implying I'll stay in research.
>You will if you start your own company or get hired as a senior ML researcher for a gigantic corp like say, Google but will take about 5-10yrs to get there even w/PhD.
Implying I haven't already had an internship at Google, or that 80% of graduates in my department don't immediately start working at one of the big tech companies.
>Google at first would pay you probably the bay area standard of 150k which is what a front end developer makes too.
Implying a front end developer has any potential for career advancement.

...