Algorithms Are The Language of Innovation
While many will debate the definition of Machine Learning (ML), very few will contest its increasing importance to those enterprises striving for a competitive edge. In its broadest form, Machine Learning is a technical discipline which aims to derive high value insights from low value and often raw data - with little to no human supervision. The defining factor being the level of human interaction that is required for learning to actually occur.
It’s important at this stage to make a distinction between Artificial Intelligence (AI) and Machine Learning which are often used interchangeably. AI is the pursuit and as yet, unrealised objective of recreating characteristics of human intelligence in machines. Machine Learning in contrast, attempts to create predictive models in the performance of specific tasks.
Despite its recent popularity Machine Learning is not actually a new science. Researchers interested in AI have for some time used pattern recognition in data, to determine if computers could learn to perform tasks - without being programmed specifically to do so. The surge in recent interest has a lot however to do with what many refer to as the Big Data Movement. The volume and complexity of enterprise data has now grown to a point where conventional data mining or analysis, performed by humans, simply isn’t effective enough. Machine Learning has effectively superseded terms like ‘data mining’, ‘business intelligence’ and ‘analytics’ as a part of the wider Big Data revolution.
The Big Data challenge has in many respects become insurmountable for humans alone. Traditional data mining comprised a team of analysts, who would pour over variables manually looking for patterns and relationships - that could potentially derive insights. Not only was this approach time consuming but it was often like trying to find a needle in a hay stack. Machine Learning in contrast, has the potential to identify data patterns in a fraction of the time it would take even the most skilled human analyst. In addition to increased velocity, Machine Learning also has the potential to identify patterns which would ultimately elude human analysis entirely.
In an enterprise context, Machine Learning is most likely to take the form of software or an application. The application or agent as it is often referred to, takes data as a low value input and in turn outputs a high value product which could comprise anything from patterns, equations and rulesets to program code or even fully formed insights. The output itself can range in composition from machine readable data, to that intelligible by humans or a combination of the two.
There are 3 widely recognised Machine Learning methods each defined by the level of requisite supervision; supervised learning, unsupervised learning and reinforcement learning. The fundamental difference between supervised and unsupervised learning is the amount of data labelling used. The former analyses labelled data (often in IO pairs) whereas the latter does not. Reinforcement in contrast, enables machines to learn using reward feedback provided by the environment.