The Unconventional Guide to Machine Learning
This is an easy-yet unconventional guide to introduce you to Machine Learning for those who are starting this journey.
This article is totally inspired by Cassie Kozyrkov.
Whenever I hear people wondering what Machine Learning is, I want them to explain this field as much as possible because, at first glance, it sounds fancy. Probably because it sounds like it, it could drive them away even further.
In other words, sometimes it sounds like it's out of reach. In this article, I'm going to explain in a friendly way what Machine Learning is.
Before I start, why do I chose the title to be Unconventional Guide? Basically, it's because I don't want to use technical or complex concepts. I'm not suggesting that complicated terms are wrong. Still, I just want to explain one of the most mentioned concepts today in the simplest possible way.
What is Machine Learning?
According to IBM website:
Machine learning is a branch of artificial intelligence (AI) and computer science which focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy.
To my understanding, Machine Learning is learning from a collection of examples on how to perform a particular task. In other words, it's the art of labeling stuff: Just tell the computer what you want to predict, and it will find the patterns given the data.
Indeed, this wave/trend of machine learning appears to be new to most people entering this field. Still, most of the algorithms, mathematical concepts, and statistical foundations related to this domain have been there surfing on books, papers, and the internet all these years.
Nowadays, since computers run and solve more complex problems than humans, they are far better at solving well-defined problems. Besides, performance is much better than a human when a large amount of data is involved. With this in mind, we can expect the machines to come up with better solutions by recognizing the patterns in the databases.
Regarding the art of labeling, let me show you what I mean.
If you ask a Dog what is this:
Somehow the Dog will try to tell you that it is a bone, right? But what if you ask the same Dog what is this:
Well, I guess the Dog won't know soon, and even that puppy won't understand what it is.
Now, let's try with you!
Try to make your best guess of what is this:
Unquestionably you can identify it as a dragon. Still, somehow your brain made something complex computation to quickly identify it and classify it as a dragon. By now, you can have an idea of the dragon size and possibly where you could expect to find one, either because of having read or heard a story about dragons.
Bearing that in mind, you can learn what a dragon is by reading and searching through books or stories, or you could understand what a dragon is by examples.
Yes, just examples, and it should be relatively simple, right?
In fact, how do babies learn faster? Suppose you have kids, nephews, or whatever. In that case, you might be astonished at how such a small living being can be able to identify how a dog "woof," how a horse "neigh," or how a pig "oinks" so easy, and it's because of examples!
The same logic should apply to computers because when you start teaching them by examples, it becomes easier for the machine to predict what you are trying to say.
Can Machine Learning be categorized?
Absolutely! Machine learning has many algorithms and subdivisions, but essentially the two most popular types are:
Supervised is when you use a dataset containing sample training with the appropriate associated tags. Labeled inputs are given to the algorithm to expect specific outputs. Per tag, we know which output variable to predict. Within supervised learning, there are two kinds:
Regression: Regression types simply predict a regular result that is not categorized. In short, expecting a number is a regression case. A typical example could be predicting house prices given characteristics such as neighborhood, size, number of rooms, etc.
Classification: The results are classified into Machine Learning classification types, meaning the target variable is predicted on a discrete scale. A classic example would be whether an email is a spam or not.
Popular Supervised Machine Learning Algorithms:
- Linear Regression
- Decision Tree
- Logistic Regression
- Random Forest
Unsupervised is when you use algorithms to draw inferences from datasets consisting of input data without labeled answers. In fact, there is no "human" involved in the learning feedback loop, and there is no label telling the computer when it is right or when it has made an error so it can correct itself.
Popular Unsupervised Machine Learning Algorithms:
- Hierarchical Clustering
This article intends to give you a friendly introduction and not-so-conventional machine learning guide. Keep in mind that this field is related to the art of labeling things correctly to predict the target variable you want or to detect or find specific patterns.
I didn't go into many terms or algorithms because I didn't want to overload you with information. If you have any questions, please do not hesitate to ask!
P.S. In case you're wondering what I didn't cover, some of the subjects might be:
- Reinforcement Learning
- Deep Learning
- Dimensionality Reduction
- Neural Networks
Kozyrkov, Cassie. “The Simplest Explanation of Machine Learning You'll Ever Read.” Hacker Noon, 24 May 2018, hackernoon.com/the-simplest-explanation-of-...