Computers Learning Things

By James

Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning Machine learning Deep Learning. Nowadays, this stuff is just floating around on the internet, from random social media posts to this article that you are reading right now. The meaning that many have accepted of Machine Learning (ML) and Deep Learning (DL) is just "computers learning things," but it is so SO much more.

Quick Overview

In short, Machine Learning and Deep Learning are algorithms that learn from the data fed to them, where they can find patterns in their data, and predict an output like autocomplete basically (even though it may be a little weird sometimes). When you type in text on a phone, there's some suggestions on top of the keyboard. Those suggestions are ML, in a nutshell, predictions from inputs and patterns in data. Autocomplete finds patterns in its data and outputs its suggestions to you based on the text that you have typed. Now, Deep Learning is a little more complicated and involves some more things in between. Its algorithms also include complex neural networks that imitate how the human brain functions to find patterns in large amounts of unstructured data. In the end, though, its primary purpose is still to generate an output based on input and the data fed to it.

Machine Learning

Ok, so I just said that Machine Learning was algorithms predicting outputs from data and inputs. Great, so that's it? Not exactly. Yes, it's true that Machine Learning does all the stuff above, but what the algorithms are is something that's more complex than just "Learning Things." The algorithms can take on many forms, but one that is commonly found is Linear Regression. Linear Regression needs two variables for the most part, one variable representing what you want to predict, and another that you are using to predict the other variable's value. Basically, one value predicts another. Keep in mind that there needs to be data for both variables for the algorithm to learn what it's trying to predict. Once you give your algorithm data, with Linear Regression, the algorithm will try to draw a line that best fits the data and the variables that it is trying to predict, finding a pattern in your data and then being able to create an output from it with an input. Linear Regression though, assumes that all data is in numeric form, so it is most suitable for usage with numerical data.

Deep Learning

Well, Deep Learning is also Machine Learning, kind of. Just more complex. It's on the more interesting side of algorithms. Machine Learning primarily finds patterns in data, and creates outputs based on inputs. Deep Learning is more so using artificial neural networks, mimicking the structures of the human brain in a network of algorithms, or artificial neurons and layers used to find complex patterns in large amounts of data. It's a bit like how we can make out patterns in stuff that a line or variables can't comprehend, so its accuracy and performance tend to be better than Machine Learning when using large amounts of complex data. In an Artificial Neural Network, there are input layers responsible for the inputs that the Deep Learning algorithm receives, hidden layers that the inputs go through and are modified to create an output, and the output layer, where an output is produced. An input is passed through each layer and their neurons, before an output is produced. Essentially, more layers in an artificial neural network means more depth, or a "deeper" neural network. The thing about Deep Neural Networks and Deep Learning is that even though they perform better than Machine Learning in some cases, they do require a lot of computing power and training data, so there are still a few downsides to Deep Learning. (But there are free computing resources in places like Kaggle and Colab and tons of data out there, so beats me)