Understanding Machine Learning and Some of the Maths Behind It

Machine learning refers to any computer algorithm which has been trained with real data to build predictive models. It is a subfield of artificial intelligence which relies on mathematical tools including linear regressions and Bayesian statistics. 

Every machine learner enthusiast begins with linear regression algorithms, models which assume a linear relationship between input variable (X) and output variable (Y). For illustrative purposes imagine that you have collected the heights and weights of 50 girls. This “training data” is then plotted into a graph (heights on the X axis and weights on the Y axis) and a line of best fit (y=mx+c) can be added. The aim is to be able to use this line to predict the weight of other girls not included in the training set (new data) given their known height. The linear regression method trains the algorithm based on the best values obtained by the line of best fit to predict further data. 

Bayesian statistics essentially calculates the probability of an expected event outcome by taking prior knowledge (probability) into account. While a traditional statistic example would be “what is the probability of rain today?” a Bayesian probability would be “what is the probability of rain given that it was sunny yesterday”. As such the calculated probability changes as new information is fed in. The Bayesian approach is not appropriate for all machine learning models but particularly relevant in data with high levels of uncertainty or databases with few training data.

Featured Courses

The CPD accredited courses are carefully crafted to help you gain in-depth knowledge on a topic of your interest.

Two main types of machine learning exist:

  • Supervised learning: the used training data is known and used for the output we are attempting to predict. As such the accuracy of our model can be investigated as we can compare it against a set of “known correct answers”. Supervised models may either be a classification type (output variable is a category for example “a risk” or “not at risk” or a regression type (output variable is a continuous value “next fashion trends”). 
  • Unsupervised learning: the system here does not predict the correct output but analyses the data to unveil unseen patterns in the data we contain. For example, this would allow the clustering (classification in groups based on similarity) of movies based on their plot summaries and actors.  Unsupervised models are either in clusters (grouping of data based on a shared characteristic “insect type”) or in association (people who listen to Camila Cabello tend to listen to Ariana Grande’s pop songs).