Machine Learning is a Broad Subject

Technically speaking, Machine Learning involves a computer running an algorithm in order to learn something, and this algorithm could be diverse in terms of autonomy, the variability of data it can handle and the predictions it can make.

It could be, in the simplest form, an algorithm derived by a data scientist looking at a specific dataset and looking for specific patterns to learn over time, or at the other extreme it could be a computer model of the human brain (Artificial Neural Network) learning from any set of diverse data thrown at it.

Machine Learning that runs specific data-science derived algorithms, although they are certainly learning, are deemed low level Artificial Intelligence as they are looking for specific patterns in specific data and are usually continually finely tuned by human operators - therefore offer relatively low levels of autonomy and variability.

On the other hand, Machine Learning that uses Artificial Neural Networks is the most high level Artificial Intelligence approach, as the system is not only more powerful in terms of being able handle and link extremely diverse data (the same neural network can be used on all types of data), but is able to function with very little (or no) input from a human observer. Artificial Neural Networks are made up of Artificial Neurons that combine in layers (input layer, hidden layer(s) and output layer) to form a network of neural connections that model the basics of how the human brain works. These ANNs are modelled using mathematical algorithms that have the sole aim of working out how an output is achieved from a set of inputs - as quickly as possible. Artificial Neurons are linked to each other using a weighting system. They each receive inputs, process them, change their state, and then output to the next neuron using different degrees of influence, much like the brain does. The process is repeated many times until the system has been trained and primed ready to accept subject data to predict from.

The Machine Learning Process

Whatever type of Machine Learning is employed, whether it's a straightforward statistical algorithm or an Artificial Neural Network, numerical data (generally normalised and converted from alphabetical data if required) is utilised as the input and Machine Learning processes use this data to learn from using three approaches, in order to link the data within, assess the relationships, and come up with an output.

Types of Learning

Supervised, Unsupervised and Reinforcement Learning

When an AI system is learning, the term is Training and in Machine Learning it happens in one of three ways:

Supervised Learning

In supervised learning - The most common form of AI learning - AI is supplied with data where we know what the inputs and output fields are and the AI uses Machine Learning to link known inputs with known outputs. e.g. in a very simple case we may have inputs of "Age", "Post Code", "Education Level", "Credit Score" and simple outputs of "Good Customer" and "Bad Customer" based on our historical data. AI would learn the links between all these data so that when we pass it information of a potential customer it will be able to tell us if this customer is likely to be "Good" or "Bad". In reality, you would likely have dozens, hundreds or even thousands of inputs from which the AI will learn from - which is why it's so powerful - making links that a human or traditional computer program would simply not be able to do in the same time frame.

Unsupervised Learning

Here, AI is supplied with data as above however it doesn't know what fields make up the inputs or outputs - it is up to Machine Learning to learn these combinations by e.g. working out the links between all fields and looking for correlations.

Reinforcement Learning - also known as Reward-based Learning

Here AI again learns through Machine Learning, however it learns by rewarding itself if it gets a particular action correct. It does this by firing off an action and then comparing the environmental state before and after the action - this type of learning has been used to successfully train AI systems to play computer games by vision alone, based on learning thousands of screen actions performed by a human player.

Machine Learning Software

Fennaio designs, develops and provides Machine Learning software than can be used for all Artificial Intelligence systems.

We have available off-the-shelf turnkey Machine Learning/AI software and can create bespoke software solutions where required.

Please get in touch with us to discuss your Machine Learning requirements.