0
$\begingroup$

I quote the well known definition:

A computer program is said to learn from experience E with respect to some class of tasks T and some performance measure P, if its performance on T, as measured by P, improves with experience E.

I have a hard time to digest what this definition exactly means.

Specifically, what with experience E means? Does it mean with more data?

Moreover, improves means that we compare with something. In that case, we compare the performance P with what?

$\endgroup$

1 Answer 1

2
$\begingroup$

Specifically, what with experience E means? Does it mean with more data?

You can think of an experience as information. These are observations of any real world system outside of the program. You can also view it as a reduction in entropy.

Moreover, improves means that we compare with something. In that case, we compare the performance P with what?

See the original quote: "improves with experience E." The word 'with' here is a common way to establish a mathematical relationship between P and T. It can be interpreted as (greater E) implies (greater P). This specifically though means that P is monotone increasing with respect to E, but it's also possible that he means P is only asymptotically increasing, which has a more complicated definition.

What Mitchell is saying overall can be interpreted simply as: when we decrease entropy (by making observations), the performance metric increases.

In my opinion, this is not really a rigorous definition of machine learning. It is just an informal description that fits a number of possible definitions of machine learning.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .