What is Machine Learning(ML)?

Today i’m going to explain a topic that is increasing in interest around the world and how it possibly applies to your career or company either now or in the future.

Machine Learning

Machine learning is a subset of artificial intelligence that focuses on creating applications that learn from a dataset and improve accuracy over time without being programmed to do so.

The way traditional programming works is you write down a set of rules that will give a result depending on the input/question that the user inserts. In other words if i type into my calculator an addition operation like 1 + 1, the calculator has been programmed to work on addition in a specific way and therefore the result will be 2.

Machine learning takes a different approach instead of writing the rules, the machine writes the rules for itself based on the inputs and results it is given. For example if you give the machine pictures of cats with the breed of the cats and pictures of dogs with the breed of the dogs, it should be able to distinguish either cat or dog if it is ever asked to do so. It should figure out for itself what a dog or a cat is. It writes its own set of rules of what dogs look like and how cats look like.

How Machine Learning works

Let’s dive a little deeper into how it works. The goal is to build a machine learning model/application.
There are 4 steps involved when an ML model is being created.

1. Select a dataset to Train

A dataset is usually a large collection of information like patient records or football club statistics. This dataset is what the model will use to solve the problem it’s designed for.
There are labelled datasets which has labels like cat or dog, breed etc. and there’s also unlabelled datasets of which the machine will classify on it’s own.

Preparation of this data is critical because it is what the ML model will use to produce results. So all the data is checked for duplicate values, biases just to make sure the model gives correct information at the end.

After preparation you split the dataset into a training set and a test set. These sets are different because if you train the model on the whole set it will produced biased results. We’re trying to avoid a scenario where the model crams those specific pictures of cats and dogs and then when given a new picture of a dog gives an error because it didn’t recognise the new picture.

Therefore this is the most important step in the process.

2. Choose an algorithm to work on the training set

An algorithm is a set of statistical steps. An algorithm is chosen based on if your dataset is labelled or unlabelled.
There are different kinds of algorithms for the different datasets:

For labelled datasets there’s regression algorithms, Decision Trees and Instance-based algorithms

For unlabelled datasets there’s Clustering algorithms, Association algorithms and Neural Networks.

There is a lot to cover for each of these i’ll expand on them in future blogs

3. Train the algorithm to create the model

This is where the magic happens. You train the algorithm on the training dataset. It is a repetitive process so depending on the size of your dataset it could take minutes, days, weeks and even years. You typically run variables through the algorithm until the algorithm yields a correct results. And this result is the machine learning model.

Side note: The words algorithm and model are used interchangeably especially by the professionals.

4. Using and improving the model

Now we can use the model on the test dataset and hopefully it improves in accuracy over time. So for our example of cats and dogs, the more pictures it takes in, the better it should get at recognising them.

Applications of Machine Learning

I’m pretty sure you’ve seen examples of machine learning but have never noticed them.

Digital Assistants like Siri, Alexa, Google Assistant use machine learning to change text to voice or voice to text all while understanding the instructions you give them.

Recommendation systems like the ones in Netflix, YouTube give you a better experience of the platform because they know what you are interested in and try their best to give you interesting videos that you might like.

Email spam detection is another one. Have you ever realised that those unnecessary emails don’t reach your main inbox. That is because there’s a machine learning model that has been tasked to separate real emails from fake ones.

Conclusion

The 4 steps when to build a machine learning model are selecting a dataset, choosing an algorithm, training the algorithm and improving the algorithm

We have just touched the surface of this topic but i hope you understand now that the world we live in is evolving and getting more efficient because of tools like machine learning and artificial intelligence and it is touching every aspect of life whether it is medicine, sports, astrophysics or the automobile industry.

Always keep learning.