Is training data used exclusively in machine learning

Question

[ATCG]

Hi all, I have a very naive question: Is a method using a machine learning approach if there is training data involved? Or put differently, is "training data" exclusively used in machine learning? I found this definition for training data: Training data is a collection of labeled information that's used to build a machine learning model. Thanks!

Answer 1

[Kevin Blighe]

No, 'training data' it is a term [and strategy] that has been used for decades. There's nothing new about anything in Machine Learning, apart from it being a marketing strategy to attract funding.

Kevin

Edit: the traditional study design would be:

1. pilot phase
2. discovery phase
3. validation phase

Training data may be used in step 2.

[Ram]

I'd go as far as to say that "training data" is mathematical induction applied to datasets. The term has become largely associated with ML though, so to answer OP's question, one can assume that if training data is involved, ML is also involved. But the other way around - that training data is not used outside of ML - might not always be the case (although it probably is true in most cases)

[i.sudbery]

Machine Learning is not itself new - the term as we understand it was coined in 1959 and had its first heyday in the 70s.

[Jeremy Leipzig]

Here is an article from 1964 that uses the term "training data" in the same sense we understand it today. So I agree the term is inextricably linked to machine learning but neither is new.

Answer 2

[Istvan Albert]

Is fitting a straight line machine learning?

We use the existing data to "train" the best fit, and we can "predict" (extrapolate) from that line.

As for a better definition:

https://en.wikipedia.org/wiki/Machine_learning

Tom M. Mitchell provided a widely quoted, more formal definition of the algorithms studied in the machine learning field:

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

thus machine learning is different from a simple formula in that in ML more data will improve the performance.

[i.sudbery]

Is fitting a straight line machine learning?

Yes. The more data you have, the better the overall fit of the line will be (the smaller the standard errors).