Recap

Let’s do a quick recap of everything we have learnt on Machine Learning.

Machine Learning is the science of teaching computers to learn from data.

“Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed.”

Arthur Samuel

The idea is that there are generic algorithms that can generate a model, or provide insights, out of a set of data without the need to write any custom code specific to the problem.

Below, you will find two articles of interest that will help you if you wish to learn more:

Introduction to machine learning: Click here to access.

AI in Finance: Cutting Through the Hype (With Case Studies): Click here to access.

bubble diagram: machine learning is divided to supervised, unsupervised and reinforcement learning

| Image: Machine Learning bubble diagram |

Unsupervised Learning finds naturally occurring patterns in the data. There are no “answers” provided in advance.

Example (clustering):

A local authority uses the data (attributes) for the residents in its area to find out groups (clusters) of residents with similar attributes. This allows the local authority to better distribute its budget to services relevant to the different groups of residents.

Example (association rules/market basket analysis):

A retailer uses the transaction data (what each customer bought) to establish links between products – which products are bought together / not bought together on a consistent basis. This allows the retailer to plan promotions and to change prices knowing which products would be affected each time.

In Supervised Learning, the training data provided as input to the algorithm are

Known inputs and known outputs
Known attributes and known labels (classes)

The algorithm has the data and the correct answers and has to find the formulae or rules in order to provide the answers to as yet unknown data.

Example (classification):

A bank uses data (known attributes) for its existing customers and their attitude to investment risk (known classes/labels) to build a classifier. Once this is done, the bank can automatically assign a new customer, based on their attributes, to the relevant class, and offer them suitable investment products.

Example (regression):

A retailer uses past prices (known inputs) and corresponding sales (known outputs) to build a predictive model. Using the model, the retailer would be able to calculate the expected sales for any set of prices.

Reinforcement Learning is a more advanced category; the learning system (agent) needs to learn to make specific decisions. The agent:

observes the environment to which it is exposed
selects and performs actions aimed at maximum utility over time
gets rewards for actions contributing to utility
gets penalties for actions detrimental to utility.

Thus, the system learns the best strategy (policy) based on trial and error and past experience.

Example (DeepMind’s AlphaGo):

The system learned the winning policy at the game of Go by analysing millions of games and then playing against itself.  At the championship of Go in 2017, AlphaGo beat the human world champion by applying the policy it had learned earlier.

That is the end of our recap. It is time for a quiz to see if you have grasped the main concepts of machine learning. Good luck!

When you’re ready, select the vertical Course button to continue.