An Introduction to Machine Learning

‘Machine learning is the subfield of computer science that gives computers the ability to learn without being explicitly programmed.’

To translate that into everyday terms, it’s the technological equivalent of taking stabilisers from your child’s bike and watching them wobble off down the garden, quickly learning, correcting and eventually steering their own course. But while they’ve been initially programmed by practice, a safe pair of parental hands and a not-inconsiderable-dollop-of-bravery, machine learning is a series of techniques which, once programmed, lets the technology infer, inform and predict autonomously, generating its own insights and effectively educating itself as it does.

While the mathematics that underpins machine learning are fiendishly complicated, the premise is simple. Using a series of instructions, or algorithms, technology is programmed to make decisions automatically using existing patterns to drive its thinking. By collecting and preparing the data for input, we can then choose – or develop – the algorithm needed, which is then evaluated and further refined if necessary.
The algorithms themselves can be broadly divided into three fields. The first, ‘supervised learning,’ uses an input > output process, with the algorithm generating a likely output based on the given information. The more information that is input, the more accurate the output, so a Spotify user new to the service, for example, can expect their recommended playlist – the output  – to increase as they themselves select more music – or input.

The second is ‘unsupervised learning,’ in which there is data input but no targeted output. Instead, machine learning is put to work by analysing and modelling the underlying distribution, thus learning more about the data. Without a specific aim in mind, the results of unsupervised learning broadly relate to trends, so it’s used most often in targeted marketing, helping companies select and appeal to appropriate customer bases.

The third field is reinforced learning, where through trial and error, the technology learns to do something better. Effectively using a feedback loop, this is seen most often in fields such as manufacturing, where it helps the precision required of robots to perform a task to improve, or the spatial awareness needed to correctly store and stack items. Reinforced learning is also critical in investment, whereby the algorithm learns to pick an optimal trading strategy.

The most familiar example of machine learning is Google, whose need for rapid response technology in the face of vast data is unparalleled. Handling up to 40,000 search queries a second, it was the first search engine to introduce machine learning into the fold in 2015, with its RankBrain algorithm coming on board to help field queries. When you type something into its search field, machine learning is working away in the background, generating suggestions as you type, mining its own knowledge to create predictions. Another example is the spam filter in your email. From learning past patterns, your email will become more adept at correctly identifying what is, and isn’t, unwanted, placing each piece of email into the correct folder automatically.

With its vast applications and ever-expanding science, the field of machine learning continues to be mined for its almost limitless potential, as it helps both communities and commerce push back their previous limitations when it comes to crunching big numbers, generating insight and providing superlative predictive capacity.