Is "machine learning" a real field or just computer science people discovering applied mathematics? I studied some basics, and most of it boiled down to statistics, algorithms, some basic numerical methods, optimization, and a few little cool things people discovered 60+ years ago. A guy at my university publishes stuff that boils down to combining artificial neural networks (70+ years old) and wavelet transforms (30+ years old). Can anyone tell me what is the fuss about?
It's basically statistics and optimization.There is almost no computer science involved except notions of computational cost.
The fuss is about discovering what works best for your application. There are no recipes. So if you want to apply it one day, you need to take educated guesses and experiment. You just need to know how and what educated guesses you might take.
If you start doing k-means on a swiss roll, you're a moron.
>>8114040
I wonder why we don't have people do this stuff in applied statistics departments instead. A lot of people I've seen trying to work in this field seem to be lacking a very solid foundation of statistics.
>>8114050
where I am, it's distributed mostly between computer vision labs, robotics labs, mathematics and CS-related labs.
I agree that reading papers can be a bit unsettling if you lack in statistics. I thought I was ok but when I had to read a few papers I realized I was lacking as well.
Does anyone know any good books/courses/videos I can use to get started with machine learning?
>>8114133
https://www.youtube.com/watch?v=UzxYlbK2c7E&list=PLA89DCFA6ADACE599
Use coursera for Ng class not YT
"Machine Learning" started in the 1970s. It only recently became "new again" with interest in consumer analytics.
Data science is truly a bubble field if I have ever seen one. That entire career choice will probably be replaced by some algorithm written at MIT.
What can I actually do if I learn everything about machine learning? How do I become involved in a useful application of this skill?
>>8114296
>It only recently became "new again" with increased compute power.
FTFY
>>8114035
its not really statistics, the only pure statistic i've come across is confidence interval.
the rest is on how to define the representations of the hypothesis space and the search algorithm to find the best one. you should have more programming background / engineering / cs than statistics. statistics is about having lots of low dimentional data and induce a fact. Machine learning is high dimentional data and about the theory of learning and representation and structure of knowlegde
>>8114035
AI is a huge meme. Semantics 101.
>>8114035
No, you're right.
>>8114035
why is le black swan man related to AI?
>>8114851
why is le swag black man related to AI?
In CS you basically have stuff that works but is boring, and stuff whose underlying theory is interesting but which doesn't do well in practise.
Machine learning is of the first kind
>>8114860
AI, predictive brain model, prediction, falsification.
>>8114860
????????
have you even read his book
>>8114040
>There are no recipes.
I think there are a lot of standard recipes. They have similar flavors, with a little bit of variation in the 'spices'.
Excellent post though anon.
>>8114296
I have a book right here from the 90's about machine learning, in it's first segment it explains how not only will it become more viable in the future when computation power increases, it will also become more easily researched, stop being an idiot.
>>8115069
This. Variations on standard recipes is the name of the game with ML.
>>8115391
There were actually a number of mathematical/theoretical breakthroughs that really brought things forward too. Even though we knew about neutral networks a long time ago it wasn't until much more recently that they became practical due to a breakthrough in the way we implement the algorithm. Similarly deep networks had another more theoretical problem that wasn't resolved until about a decade ago.
Only in the last few years did people start creating all sorts of frameworks so now productivity is really taking off. There's a lot to look forward to in ML.
>>8114035
Based NNT
