Machine learning 01

This weekend I have been listening to Science podcasts and reading a book about a Neurosurgeon’s experience on the ward. I came across the topic of machine learning and became completely absorbed by it. There goes my evenings and spare time! As I learn more about this topic, I thought to write down my learnings and share via here. At the top will be a short summary.

Summary

• Machine learning is all about the data. The quality of the data and the amount of data you have. The broader the view, the more likely that you’ll find something interesting.
• There are many examples of machine learning. Google search engine. Recommendations on Netflix. Predicting the weather.
• Machine learning originated in 1959 when Arthur Samuel defined it as ‘a field of study that gives computers the ability to learn without being explicitly programmed on what to do”. This is different as normally we tell programs what to do.
• The key to machine learning is learning. You are giving a computer the data and tools to solve a problem. The computer remembers what it learns and adds it to a database. It is important to remember that it is not an enhanced form of human learning, but it’s a new way to find patterns, make decisions and gain greater insights.
• How does it work? You have a problem. Create rules. Apply rules. Get Feedback. Either adjust the rule or add it to your memory bank.
• There’s two types of data — training data and test data.
• The training data is a small subset of the data. You can use a statistical algorithm to see how the data fits together. Any patterns can be added to a database.
• Algorithm is a method where you can predict output values. For example, an algorithm to decide whether an email is spam or not.
• Binary classification is a task of classifying data into two groups based on a classification rule.
• Hyper parameters is a parameters that is used to control the learning process.

Read more or TL:DR (too long, don’t read)

Examples of machine learning out there

There are many examples of machine learning out there at the moment, such as the search engine in google. When two individuals type in the same search query, then they are likely to get different answers as it is personally customised.

Where did Machine Learning originate?

Machine learning originated in 1959 when Arthur Samuel moved away from a computer needing to be told what to do. He coined machine learning as

a field of study that gives computers the ability to learn without being explicitly programmed on what to do.

What is the key to machine learning?

The key to machine learning is learning. You are giving a computer the data and tools to solve a problem. The computer remembers what it learns and adapts going forward.

To explain this in more detail let’s use a machine learning program that detects spam messages. So initially you create a word filter program that uses words, such as gold, lottery and winner. You then ask the program to filter out any mails with the subject lines that contain these words. The trouble with this is that you could get around this by for example, changing the o in the word to a number 0. E.g. l0ttery.

This is an example of how traditional programs use input commands that result in predefined outcomes. What you need to be doing in machine learning is changing from input commands to input data.

What type of data do you need?

In machine learning there’s two types of data: training data and test data.

What you start with is a small part of the data to use as the training data. You can use a statistical algorithm to see how the data fits together. Any patterns you see can be added to the database.

In the spam email program example, what we could use is 10,000 emails as our training set. A classified algorithm will help you split messages into spam or regular messages. This is often called Binary classification.

What you need to understand is the best classifier algorithm to use. Then you’ll tweak the hyper parameters of the algorithm until the machine makes a prediction whether the email is spam or not.

What is interesting about machine learning?

What is really interesting is how sometimes the machine makes a decision, for example, whether a message is spam or not, and how you might not know how the machine came up with that decision.