Home » Machine Learning

Machine Learning

What is Machine Learning?

With a bit of search I’ve noted two Machine Learning definitions:

From an older definition (Arthur Samuel c.a. 1950s):

From a more recent and technical definition (Tom Mitchell):

Machine Learning is a method or the ability to program a computer to learn or extract knowledge from data.

The application of machine learning nowadays becomes ubiquitous in our everyday life. From the Ads you see after a search of an item in the internet or movie recommendations from streaming applications.. etc.

In general, a machine learning problem can be classified as Supervised or Unsupervised learning.

Supervised and Unsupervised learning begins with input data and categorize.

One classic example of machine learning. Let’s say, one of your friends wants to sell a house and they want to know how much they can get. Based on a data set (e.g. area + price) collected in the area you can plot horizontal axis the size and vertical axis the price. Next is the learning algorithm, maybe insert a straight line to represent the plotted data and based on that you can predict the price of the house being sold. Or maybe use different learning algorithm approach, use of quadratic polynomials to better represent the plotted data to increase the accuracy of predicting the the house price.

Types of Machine Learning

In this section we’ll give a  high level illustration of the two most common types of machine learning –> Supervised and Unsupervised learning

In supervised type, here we feed the algorithm with training data that includes the desired result or solution called label.

Labeling is also referred to a classification task. One good example is email’s “spam filter“, email training examples will come along classified (spam or ham) and it should learn how to classify emails.

Another common example is the regression task, here it should learn how to predict. Let’s say price of house, to train feed a set of features (size, number of rooms, age, etc.)  and several examples with their predictions and labels (i.e. price).

In unsupervised type, training data are unlabled. The system should be able to learn by itself using algorithms. Some of the common algorithms are “clustering” and “anomaly detection“.

For example, you have a lot of data about you homepage visitors. You may want to classify it into groups of similar visitors. In this case you may apply “clustering” algorithm.

Challenges of Machine Learning

Looking at the first step of machine learning is to chose a learning algorithm and test it on some data.  Two things can go wrong, bad algorithm or bad data.

Bad data can be insufficient quantity or poor quality of training data. Relevance of the features will also influence the expected results. Data can also be too general that the system overfit or too simple that it underfit the results.