What is Machine Learning?

Machine learning has grown to be quite the buzzword over the past few years. Its rise in popularity does not go without justification. Innovations in machine learning could drastically change our lives in the coming future, but to understand how machine learning could accomplish such a task, we must first understand machine learning itself.

Machine learning is a subset of artificial intelligence. Computer systems are trained using data and mathematical algorithms to generate models. Then, they use these models to progressively improve their performance on a specific task that would be otherwise difficult to explicitly program.

And that’s okay; this field encompasses a lot. By the end of this post, you will understand what machine learning is and be able to derive possible use cases for yourself.

Machine learning wouldn’t exist without data. With the decline in storage cost, it is evident why machine learning is becoming so popular. There are numerous ways to generate data for machine learning. The data could come from historical records, manual input, or dynamically extracted through some form of live-stream. The key here is to use the correct data for the problem you are trying to solve because it directly impacts the results that are capable of being derived from it. So before ever getting started with machine learning, it is vital to devise a well thought-out machine learning problem.

It’s the backbone of any machine learning project. A machine learning problem is versed in such a way that a computer can learn from experience (E) concerning some task (T) and some performance measure (P), if its performance on T, as measured by P, improves with E.

If you want to build a program that can predict the electricity usage for a specific city (task). You can take historical data surrounding electricity consumption for that city and run it through machine learning algorithms (experience). Once it has successfully learned from the data, it will become more accurate at forecasting future electricity consumption (performance measure).

The choice of the performance measure is critical. The performance metric enables us to see if the computer is learning or not. The performance metric should be steadily improving for the most part. In the case of our example, we would probably use something like the Mean Absolute Percentage Error (MAPE) for the value of the forecasted consumption versus the actual electricity consumption as our performance metric.

Machine learning does not always lead to perfect predictions but instead can make great guesses based on the data it has consumed and the models it has generated (far greater than humanly possible). It will then progressively improve its models and performance.

Types of “Super Smart Robots”

Machine learning can be generally categorized into three different types based on how each one uses and perceives data.

The first and most prominently used type is called supervised learning. Supervised learning involves the input of labeled data that directly correlates with an output. In supervised learning, the computer receives the inputs and outputs needed for the model from a “teacher.” For example, we want to be able to predict the height of children once they grow up using machine learning. We would likely take variables from historical data such as childhood height, weight, foot size, wingspan, and gender into account. These inputs are known, and by using historical evidence, the output of the eventual adult height is also known. Therefore, we use a “function-like” approach through machine learning to teach the computer how to predict the future adult height of children based on these variables.

The next type is unsupervised learning; it involves data that isn’t labeled. Unlike supervised learning, the meaning of the output is not directly distinguishable from the data. Therefore, it is the job of the machine and the model to find the correlations, patterns, or relationships inside the data and deliver those insights as outputs. It becomes difficult to come up with metrics for how well an unsupervised learning algorithm is doing. “Performance” is often subjective and specific to each use case. This type of machine learning works well for categorizing information that may not be fully understood. For example, Amazon uses an unsupervised learning technique called association to discover rules that describe large portions of their purchase data to recommend additional products to people.

The last type of machine learning is the most similar to how humans learn called reinforcement learning. Reinforcement learning takes a near psychological approach to incentivize the machine with a “reward” type metric. The computer is trained to make decisions based on positive and negative feedback. It must educate itself about its environment through continuous trial and error. The setting is represented as data itself and the only feedback given typically comes in the form of a “scoreboard.” For example, reinforcement learning can be applied to such things as simulating video games where it’s possible to provide positive and negative feedback based on in-game success or failure. The “score” directly drives the way the machine makes correlations between the data. A lousy score would tell the machine to perform different actions in hopes of improving the score. Therefore, the key to this type of machine learning is that the model is based entirely on the experiences in response to the reward metric.

By this point, you should have a basic understanding of what machine learning is. For the most part, its the process of teaching machines (typically computers) how to learn by giving them a series of instructions that help them develop logic on their own by providing them with the access to the data that you want them to understand. We will continue to dive further into understanding the technical details surrounding the interworkings of machine learning in later blog posts. For now, you should be able to finally answer the infamous question: “What is machine learning?”

Related posts:

A follower of her writing, his engaging comments left her intrigued. In a moment devoid of rationality, she’d agreed to meet. The Pinot must’ve been spiked that night. There was a renaissance in his…

Plumbing repair issues are an inevitable part of owning a home. Some plumbing problems arise from the natural deterioration of pipes and fittings or from the everyday usage and neglect of plumbing…

Being a parent is one of the hardest jobs in the world for so many reasons. Raising children is an extremely difficult job in and of itself, but when you factor in the scrutiny and criticism from…