What is Machine Learning?
What is Machine Learning? Machine learning can be defined as the study of computer algorithms through experience, as well as the use of data. It can be classified as a type of artificial intelligence. Many companies utilize machine learning, and to name a few: Amazon, Google, Microsoft, and IBM. So what is the allure of machine learning? Why do big name companies use machine learning for their products?
Machine Learning Is…
Machine Learning algorithms are fed sample data, also known as “training data” in order to make predictions or decisions automatically, without someone specifically telling them what to do. Think of it like training a dog, or potty training a toddler. With all three instances, the subject is shown an example of how they need to act in order to learn the desired skill or behavior. If you are teaching your dog to sit down, you can help gently set their body down to teach them what “sitting down” is, then reward them with a treat. Continuing to do so so can result in your dog learning what sitting down is, and will sometimes sit down without being asked, since they know they will be rewarded/and it is what is seen as “correct” behavior. The same with a child being potty trained. We teach children how to sit down and use the toilet at a young age by continuously taking them to the potty when they have to go. This is seen as the “correct” behavior, so we continue to do so. We know that when faced with the need to go, our automatic response would be to find an available bathroom and use the toilet. The common factor with machine learning, potty training a child, and training a dog to do tricks is: experience. For most instances, you can’t master something right away. You need to continuously perform the action until it becomes almost second nature. In the case of machine learning, the machine trains with the sample data until it recognizes the patterns, algorithms, etc, so it will be able to perform these actions on its own. This can be extremely handy in the technical field, for computers/machines to perform actions without being told to do so.
Machine Learning Approaches
There are several approaches to machine learning, but they are mainly put into three categories: supervised learning, unsupervised learning, and reinforcement learning.
Supervised Learning
First is Supervised Learning. As the name implies, it is when the machine is fed training data which is marked with all the correct answers already. We call this “supervised learning” because this process can be compared to a teacher or tutor guiding their students down a straight path to the correct answer. Supervised Learning is usually done in the context of classification, for example, mapping input and output labels. Data Scientists use supervised learning because it is already giving the answers/necessary information or steps for the machine to make, but they still need to keep an eye on the algorithm to ensure that the learning process is smooth and any insights remain true. Common algorithms in supervised learning are: logistic regression, naive bayes, support vector machines, artificial neural networks, and random forests.
Unsupervised Learning
Next is unsupervised learning. Unsupervised Learning is a machine learning approach that aims to cluster unlabelled datasets. The algorithm has the potential to discover hidden patterns, without any human supervision. Unsupervised learning models typically are used for three main objectives, being clustering, association, and dimensionality reduction.
> Clustering is a data mining technique used for grouping any unlabelled data based on either similarities or differences. It is similar to looking through a pile of fresh laundry, and pairing socks. Our brains search for similarities in style, color, length, and we pair them together. Clustering could be used for market segmentation and or image compression.
> Association is an unsupervised learning method that uses certain criterias to find links between variables in a dataset. An example would be when you’re shopping online and see sections called something like “Customers Who Bought this Item also Bought…” or “People in Your Area Also Like…”
> Dimensionality Reduction is an unsupervised learning method that is typically used when the dataset is too high. It can reduce the size of the dataset to a smaller, more manageable size, but can still preserve the data integrity. This technique can be used by auto-encoders to remove noise from data visualizations to improve image quality.
Reinforcement Learning
Lastly, Reinforcement Learning. Reinforcement Learning is when a machine uses trial and error in order to find the solution to a problem. When the machine makes the correct decision, they are rewarded. If they make an error, they are penalized. This way of learning is very much like the carrot and stick approach. The carrot and stick approach is a metaphor for using a combination of reward and punishment in order to induce a certain behavior. The machine wants to maximize their reward as much as they can, which in the process, teaches them what the correct answers and predictions are.
Why Do We Use Machine Learning?
So, why is Machine Learning so important? Why do we bother using machine learning in our projects? The answer is: data. Data is extremely valuable — especially to industries such as manufacturing, retail, healthcare, travel, financial services, energy, and utilities. We typically use Machine learning to filter through data — to analyze and interpret patterns found in the datasets. Knowing what the people want, who your target audience is, etc, is very important and can be the key difference between keeping up with your competitors or falling behind. It would be tedious for human beings to sift through data all day, so we have taught our machines algorithms so they can find the patterns/irregularities for us. It is because of machine learning that we are given recommendations in apps we use on a regular basis, such as Amazon, who will recommend items to you based on what you have been browsing/ordering in the past. Or Netflix, which gives you recommendations on what to watch next based on your browsing history. Self-driving cars also utilize machine learning to figure out rules and patterns. Machine learning can be very helpful and can make a big difference.
