Dear Non-Existent Grandfather
Does the words “Machine Learning” mean anything to you? Have you ever heard of STEM?
--
No- when I say STEM I don’t mean the stalk of a flower. STEM actually stands for Science, Technology, Engineering, and Math. It’s what your granddaughter is studying in, but to be more specific, I’ll be a software engineer in the near future.
You’ll probably never get this letter, I don’t even know who you are, but I’m writing this for my awesome school that’s taking me on this path. I’m looking forward to specializing in AR/VR but honestly- you probably have no idea what that even means. So let’s start at the basics!
What Is Machine Learning?
Does The Letters AI Sound Familiar?
Have you ever watched the movie The Terminator? It’s the movie with Arnold Schwarzenegger who’s this android assassin that delivered one of the most famous lines ever- “I’ll be back”.
So take a seat mister grandfather, because that killer android and SKYNET are examples of AI- Artificial Intelligence. Don’t worry! Today’s AI probably aren’t evil killers like them. In fact, artificial intelligence is within many different kinds of fields and holds different subsets. One of these subsets actually happen to be machine learning!
Remember: Artificial Intelligence and Machine Learning are NOT the same thing!
So What is Machine Learning Really?
As software is what directs and enables computers to work, machine learning is a kind of software that is actually able to learn from the data and statistics it’s fed in order to make future decisions and predictions.
So mister grandfather, as I don’t know how much experience you have in math, let me remind you what data and statistics are.
In this case, data is information that the computer will analyze/process, while statistics is a form of data. Statistics can encumber many things as it’s a whole branch of mathematics, but when it comes to machine learning, this statistical data is broken down into quantitative information for the computer to take in.
The computer then categorizes this information for future predictions or decisions, hence the “learning” part in “machine learning”.
The Ways The Machine Learns
That being said, there are actually many different learning methods, but two of the most common/broadest categories are supervised and unsupervised learning. Every learning method is run by a kind of algorithm, which is basically a procedure that the computer follows step-by-step.
Supervised Learning
Supervised learning is when the machine is trained by humans through pairs of input and output. The output tells the machine what the desired outcome is depending on the input. Based off this training data, it then can map future inputs and return their suggested outputs.
Two commonly known kinds of supervised learning are:
Classification: When there are a limited amount of specific values that the outputs could be.
Regression: When the outputs are any numerical value within a range.
Both of these then use algorithms to fulfil their purpose. Classification may use logistic regression while regression may use linear regression. On the other hand, there are algorithms like k-nearest neighbor (K-NN for short) and neural networks, which can be used by both classification and regression while logistic and linear regression could not.
And another kind of supervised learning would be:
Similarity Learning: Closely related to classification and regression with purpose of measuring how similar or related two objects are.
Unsupervised Learning
Unsupervised learning is different from supervised learning in a couple of ways. First off, the algorithms take in data sets consisting only of inputs- no suggest outputs. This is unlike supervised learning, where given inputs are already identified with their outputs. The algorithm within unsupervised learning finds a structure or some sort of pattern within the unlabeled input in order to classify or categorize them.
This can be done through a couple of ways:
Clustering: When the inputs are placed into clusters, or subsets, based upon certain observations. These observations are criteria that data meets, and if certain input has observations from multiple clusters, then the observations are dissimilar.
Density Estimation: Differs from clustering that finds groups in data, as density estimation summarizes how the data is distributed.
Both of these would then use algorithms to fulfil their purpose. For example, cluster analysis may use the k-Means algorithm while density estimation may use the algorithm called Kernel Density Estimation.
If this data would need to be graphed or plotted then an additional kind of unsupervised learning would be:
Visualization: Creates different kinds of plots with the data, such as scatter plots.
Common Uses for Supervised and Unsupervised Learning:
As supervised and unsupervised learning are different, they would have different uses.
An example for supervised learning is if the algorithm is given labeled images (the image would be the input and the label would be the output) then next time the algorithm is given an unlabeled image of the same kind- then it would know what label to give it.
Some reasons as to why supervised learning would be used is in the case of using historical data to predict future events that are likely to happen, like stock market fluctuations.
In the case of unsupervised learning, it is commonly used in transactional data like recommender systems. These recommender systems can help determine what to someone of a certain age range may prefer or help the bank detect a fraudulent card purchase.
And There’s Still More Kinds of Learning!
While supervised and unsupervised learning are the most broadest and well known kinds of learning, there are still more! Like reinforcement learning, which involves a software agent, so a kind of bot, that learns how to act in a certain environment. There’s also self learning, feature learning, and hybrids like semi-supervised learning and self-supervised learning.
References
Brownlee, J. (2019, November 11). 14 Different Types of Learning in Machine Learning. Retrieved January 20, 2020, from https://machinelearningmastery.com/types-of-learning-in-machine-learning/
Machine learning. (2020, January 23). Retrieved January 20, 2020, from https://en.wikipedia.org/wiki/Machine_learning
Tagliaferri, L. (2017, September 28). An Introduction to Machine Learning. Retrieved January 20, 2020, from http://www.digitalocean.com/community/tutorials/an-introduction-to-machine-learning
