Machine learning (ML) is defined as “the field of study that gives computers the ability to learn without being explicitly programmed.”
To give this more meaning, we can think of it like this: Suppose we have a computer program executes some task T, and the performance of this program in executing the task T is measured by some performance metric P. That computer program is said to learn from experience E with respect to some task T, if its performance on T, as measured by P, is improved by the experience E.
<aside> 💡 For example… suppose T represents detecting spam emails, and P is the ratio of correctly classified emails to the total (called accuracy), then an experience E could be a large set of examples of spam and non-spam emails.
</aside>
Continuing with the spam email example, the traditional approach would be:
Where we explicitly give the computer rules to follow when detecting spam emails, like keywords. The issue is that making an exhaustive and accurate set of rules is extremely complex. The machine learning approach is rather to look at a bunch of examples:
And use that as a way to decide if this new instance of an email is spam or not. Here is the updated approach:
Another place where machine learning shines is for problems that either are too complex for traditional approaches or have no knows algorithms, such as speech recognition.
Machine learning can also be used to find patterns in data that weren’t evident, by looking at what the model uses to make its predictions.
ML algorithms in general can have many characteristics: