Examples of machine learning working in the background as part of our day-to-day activities grow day by day, yet most people remain completely oblivious to the impact machine learning is having on society and on business.
This might be because they don’t really know what it is or how it works – even the experts can’t agree on a single definition for machine learning…
Can you define machine learning?
In the book “Understanding Machine Learning from Theory to Algorithms”, the authors Shai Shalev-Schwarz and Shai Ben-David refer to machine learning as “automated detection of meaningful patterns in data.”
In the popular Coursera machine learning course taught by Google Brain co-founder Andrew Ng, he quotes Arthur Samuel (widely credited as the person who coined the term ‘machine learning’ in 1959), describing machine learning as “The field of study that gives computers the ability to learn without being explicitly programmed.”
"A computer program is said to learn from experience ‘E’ with respect to some class of task ‘T’ and performance measure ‘P’, if its performance at task in ‘T’, as measured by ‘P’, improves with experience ‘E’." - Professor Tom Mitchell, Machine Learning Department, Carnegie Mellon University
As with all equations, this looks mind-boggling at first read, but let’s break it down – it’s a simple concept when you understand it.
- E = the data that is collected (Experience)
- T = the decision the software needs to make (Task)
- P = how to evaluate the results (Performance)
A good example of this is when filtering spam emails. The computer program learns by reviewing the previous emails that were labelled as spam by the user (Experience). From that, it will make a decision on whether a new email is spam or not spam (Task). Then it will evaluate the amount of errors that the program made (Performance). You see – simple!
Why we need machine learning
Machine learning is required when problems become too complex and/or there is a need for adaptivity – for example, for a task that humans routinely do, but where our own thoughts on how our bodies actually go about doing the task are not elaborate enough to extract a well-defined program, such as speech recognition or image understanding.
Machine learning is also needed when there is a task that requires analysis of large and complex data-sets. Within large and complex data-sets lies treasures of meaningful information, but their size and complexity can make deriving information from it, information that makes actual sense, an almost impossible task for a human being.
Just as humans learn in different ways, machines also learn in various different ways.
How do machines learn?
If we think of the interaction between a learner (you, or me, or a machine) and their environment (what’s around them or what is to hand), we can divide learning into two different formats, depending on how the interaction occurs – supervised learning and unsupervised learning.
Supervised learning is all about using experience to gain expertise. In a machine learning scenario, the experience (existing, labelled data) has a large information base, and when an unseen example (new, test data) appears, it applies its expertise.
Supervised learning can be further broken down into classification and regression. The main difference between the two is that with classification the output is a discrete variable (categorical), e.g., it's a spam email, whereas with regression the output is a continuous variable (numerical value), e.g. this house will sell for £396,990.
An example of a machine learning classification task would be classifying whether an email is spam or not spam. An example of a machine learning regression task would be to predict house prices.
Unsupervised learning has no distinction between the experience (existing data) and the unseen examples (new data). Instead, the learner (machine) absorbs the experience in order to summarise the data, or reveal a compressed version of it.
For example, if your company wanted to group customers together into distinct categories by clustering the data into subsets with shared similarities, your machine learning tool would use unsupervised learning. This is used today by the music streaming services, Spotify, to cluster different songs and users into different categories.
Reinforcement Learning is another method of learning, where experience is fed into the machine learning program, but the test data us not known. The learning system (called an agent) observes the environment (teacher), and then selects and performs actions. When the learner (the machine learning tool) carries out the actions that the teacher (environment) wants it to do, it is given either a reward or a penalty. The learner (machine learning tool) starts to continue carrying out the actions that gave it a positive reward, and stop those that caused a penalty – learning all by itself!
Google Deepmind’s AlphaGo is a well-known example of ‘reinforcement’ machine learning. It was the first computer program to defeat a Go world champion player in October 2015.
Want to learn more about machine learning?
The world today has more data, and more complex data, than ever before. But humans are humans – our minds can only take us so far. Machine learning is in its early stages, but it’s set to revolutionise how you conduct ecommerce, do business, and ultimately deliver services, products, and experiences to your customers.
Our AI and machine learning experts are already developing programs, tools, and new services that are changing how our customers conduct business, for the better, and they’re on hand to explain machine learning in more depth, show you real-world use cases, and brainstorm scenarios when machine learning could be of real value to your business.
