Machine Learning

Introduction

Machine learning is a technology that allows computers to learn directly from examples and experience in the form of data. Traditional approaches to programming rely on hard- coded rules, which set out how to solve a problem, step-by-step. In contrast, machine learning systems are set a task, and given a large amount of data to use as examples of how this task can be achieved or from which to detect patterns. The system then learns how best to achieve the desired output. It can be thought of as narrow AI: machine learning supports intelligent systems, which are able to learn a particular function, given a specific set of data to learn from.

Machine learning came directly from minds of the early AI crowd, and the algorithmic approaches over the years included decision tree learning, inductive logic programming, clustering, reinforcement learning, and Bayesian networks among others. As we know, none achieved the ultimate goal of General AI, and even Narrow AI was mostly out of reach with early machine learning approaches.

In some specific areas or tasks, machine learning is already able to achieve a higher level of performance than people. For other tasks, human performance remains much better than that of machine learning systems. For example, recent advances in image recognition have made these systems more accurate than ever before. In one image labelling challenge, the accuracy of machine learning has increased from 72% in 2010, to 96% in 2015, surpassing human accuracy at this task. However, human-level performance at visual recognition in more general terms remains considerably higher than these systems can achieve.

While not approaching the human-level intelligence which is usually associated with the term AI, the ability to learn from data increases the number and complexity of functions that machine learning systems can undertake, in comparison to traditional programming methods. Machine learning can carry out tasks of such complexity that the desired outputs could not be specified in programs based on step-by-step processes created by humans.The learning element also creates systems which can be adaptive, and continue to improve the accuracy of their results after they have been deployed9.

Machine learning lives at the intersection of computer science, statistics, and data science. It uses elements of each of these fields to process data in a way that can detect and learn from patterns, predict future activity, or make decisions.

As it turned out, one of the very best application areas for machine learning for many years was computer vision, though it still required a great deal of hand-coding to get the job done. People would go in and write hand-coded classifiers like edge detection filters so the program could identify where an object started and stopped; shape detection to determine if it had eight sides; a classifier to recognize the letters “S-T-O-P.” From all those hand-coded classifiers they would develop algorithms to make sense of the image and “learn” to determine whether it was a stop sign.

Good, but not mind-bendingly great. Especially on a foggy day when the sign isn’t perfectly visible, or a tree obscures part of it. There’s a reason computer vision and image detection didn’t come close to rivaling humans until very recently, it was too brittle and too prone to error. Time, and the right learning algorithms made all the difference.

Branches of machine learning

There are three key branches of machine learning:

supervised machine learning

In supervised machine learning, a system is trained with data that has been labelled. The labels categorise each data point into one or more groups, such as ‘apples’ or ‘oranges’. The system learns how this data – known as training data – is structured, and uses this to predict the categories of new – or ‘test’ – data.

Unsupervised machine learning

Unsupervised learning is learning without labels. It aims to detect the characteristics that make data points more or less similar to each other, for example by creating clusters and assigning data to these clusters.

Reinforcement learning

Reinforcement learning focuses on learning from experience, and lies between unsupervised and supervised learning. In a typical reinforcement learning setting, an agent10 interacts with its environment, and is given a reward function that it tries to optimise, for example the system might be rewarded for winning a game. The goal of the agent is to learn the consequences of its decisions, such as which moves were important in winning a game, and to use this learning to find strategies that maximise its rewards.

When machine learning systems are deployed, there is a key distinction between offline and online learning systems:

• Offline learning systems are trained and tested in an offline setting, and the trained models are then ‘frozen’ before being deployed to a live setting. Any subsequent training will also be performed in an offline setting, tested, and then deployed using conventional software change management methods. This approach is more common in machine learning systems that are deployed today, because it gives an opportunity for human verification of the system, before the system interacts with any user.

• Online learning systems are also trained and tested in an offline setting before deployment, but the learning algorithms continue to be applied to the trained model after deployment. This means that the performance of the system ‘in the wild’ can continue to improve in real-time in response to real-world data. It also means that there is no opportunity for human checking of the consequences of updates to the model, before users are exposed to these. For example, many email spam detection systems perform online learning in response to patterns of inbound email and user feedback on the system’s accuracy.