AI and ML White Paper Series for TSG – Unsupervised Learning
One of a series of six white papers for my client on the subject of AI and the different types of machine learning. Each acted as a companion piece for the others, ensuring that a highly complicated subject could be explained in easily digestible chunks.
What is unsupervised learning?
Unsupervised learning is where a machine learns for itself. It has to identify naturally occurring patterns within data sets on its own – there are no pre-assigned labels or scores to help it. The conclusions it draws are based on statistical methods such as clustering and modelling.
- While being fundamentally backwards-looking, it can result in new and unexpected conclusions in the data and hidden structures being revealed.
- Unlabelled data is faster and easier to obtain than the labelled variety and is quicker to input as well. Yet it can still be used to undertake complex tasks.
- It reduces the likelihood of mistakes – labelling takes, training, time and effort, and can suffer from human error.
- It’s useful for testing the efficacy of AI in a given situation.
- It generally results in less precise and slower training compared with supervised learning.
- Its very unpredictability can mean the creation of unwanted categories and conclusions, resulting in confusion and error.
- There’s a risk of learned data bias, which can become normalised over time – it can take a while for errors and unwanted categories and results to come to light.
- This means that while less time may be spent on input, more may have to be spent on interpretating the results.
Sometimes, the line between supervised and unsupervised learning is blurred by semi-supervised learning. With this, a part of the given input data has been labelled. This can reduce unpredictability, but at the expense of greater cost and time input.