Supervised Learning Versus Unsupervised Learning

How are human teams helping AI?

Machine Learning is the primary field that involves teaching AI systems to learn with data. Inside of machine learning, there are two typical sub-fields of learning. We hope that with a better understanding of these sub-fields, it will become easier for our readers to understand current developments as well as current roadblocks in the Artificial Intelligence industry.

What is unsupervised learning?

Unsupervised learning is when an AI system has a large amount of past data to work with that is not labelled in any way by human experts. Therefore, the learning occurs in an unsupervised way.

Related to this, the Machine Learning Mastery blog reports that the hitch with unsupervised learning is that there are no correct answers. What this means is that it is now the AI system’s job to find patterns and areas of significance in the data that it takes in.

More specifically, this usually takes the form of mapping the structure of said data, which involves two paths of doing so. These paths are split into the types of problems that can occur under unsupervised learning and are called: clustering and association problems.

Essentially, an example of a clustering problem could be the type of problem that happens when trying to put the customers into personas based on their demographics and purchasing behaviors.

An example of an association problem, on the other hand, appears to be more wide ranging. This is usually thought of as the type of problem that happens when a system is trying to make connections like the Amazon “because you bought, we recommend,” algorithm does.

All in all, there seems to be a wide range of risks that exist in unsupervised learning related to the fact that systems are left on their own with no past data fed into them to guide their learning.

Given the fact that the AI industry is currently in its infancy and misclassifications of data seem to be widely reported across use cases of AI systems, it does appear logical to claim that unsupervised learning is not the most recommended way to go.

What is supervised learning?

Supervised learning is when an AI system is, in fact, fed training data either based on past successes with what it is built to do or past successes from other systems or experiments with what it is built to do.

According to the Machine Learning Mastery blog, most of today’s machine learning work uses supervised learning. The DataScience.com blog concurs with this assertion.

One of the logical reasons that this seems to be true is that the process of supervised learning effectively mirrors forecasting in one key way.

Forecasting uses past and present data to predict the future of a financial market, for example.

Supervised Learning uses past and present data to predict the future outcomes that will lead to the ideal performance of an AI system related to its task.

Therefore, it could be said that supervised learning uses the theory of forecasting combined with learning algorithms to reach its aims. To review, learning algorithms are simply functions that use input data from the past and present to predict the best future outcomes or output data for an AI system.

The goal of doing this, according to the Machine Learning Mastery blog, is to eventually understand past data so well that when present data is inputted into the system, the system can predict what the output values should be.

It is true that all of this is highly theoretical and a logical percentage of error will continuously exist despite efforts to reduce it. Look no further than a 5% rate of error being lauded as a record in the space of AI transcription of human phone conversations. Error in AI systems can’t be eliminated, at least, not yet.

With this brief overview of supervised and unsupervised learning, it is hoped that we can move into pieces that show these forms of machine learning in practice, without widespread review of their foundational concepts. On top of this, it is also hoped that this can help you, the reader, to understand why supervised learning is largely being held as the industry standard of Machine Learning, today.