HN Explorer

Ask HN: What precise problem does machine learning solve?

2 points by Tomminn ↗ HN
Can someone give me a precise specification of the problem that machine learning actually solves. This would be useful so anytime someone suggests it, I can start my line of reasoning with "the problem machine learning solves is..."

Analogous example: "the problem the blockchain solves is decentralized trust."

5 comments

[ 4.0 ms ] story [ 24.1 ms ] thread
tomasNth
outputing the extcaction of the right content from the input. learning is extcacting. so "the problem machine learning solves is machine content extcaction"
pavbelshippable
"The problem the machine learning solves is interrupted pattern analysis."
bufferoverflow
Machine learning doesn't solve one problem, but thousands. It's like asking what problem does a knife solve - it can be used to cutting, stabbing, throwing, scaring, etc.

ML is used in the areas where there no known algorithms, or the algorithms are too slow or simply not good enough.

In more general terms, ML tries to provide hgh quality outputs for given inputs.

Tomminn
I think this can't be right. Machine learning is a bunch of small variations to a central algorithm. There must be one specific problem that it solves, otherwise it would be thousands of radically different algorithms. The second half of your answer is closer to what I'm looking for, but it's too vague.
bufferoverflow
ML is a bunch of different algorithms, and each of them has their own variations. Like deep learning is a general one, but then there are all kinds of topologies that result in various applications and performance.