Machine learning algorithms are responsible for the vast majority of artificial intelligence developments and applications you hear.
What is the definition of machine learning?
Machine learning algorithms use statistics to find patterns in large amounts of data. The data here covers a lot of things – numbers, words, images, clicks, etc.
Machine learning is the process that powers most of the services we use today – suggestion systems such as Netflix, YouTube and Spotify; Search engines such as Google and Baidu; Social media posts such as Facebook and Twitter; Voice aids like Siri and Alexa.
In all these cases, each platform collects as much data about you as possible – you might want to use machine learning to make a highly educated guess about what types of likes you watch, which links you click, which situations you react to – and what you learn.
Obviously, this process is quite simple: find the pattern, apply the pattern. This was largely accomplished in 1986 by Geoffrey Hinton, now known as the father of deep learning.
What is deep learning?
Deep learning is machine learning on steroids: it uses a technique that gives machines the ability to find and upgrade even the smallest patterns. This technique is called a deep neural network – it is deep because there are many, many layers of simple compute nodes that work together to mix data and deliver an ultimate result in the form of predictions.
What are neural networks?
Neural networks were uncertainly inspired by the inner workings of the human brain. Nodes are like a kind of neuron, and the web is like the brain itself. Hinton, however, published the breakthrough article at a time when neural networks were outdated. Nobody knew how to train them, so they didn’t give good results. The return of the technique took about 30 years.
What is supervised learning?
One last thing you should know: machine (and deep) learning comes in three ways: supervised, unsupervised, and reinforcement. In the most common supervised learning, data is labeled to tell the machine exactly which patterns to look at. Once you know this, think like a relaxing dog that will hunt targets. This is what you do when you press the play button on a Netflix show – you tell it to find algorithm-like shows.
What is unsupervised learning?
In unsupervised learning, data has no label. The machine only searches for patterns it can find. This is like letting a dog smell tons of different objects and dividing them into groups with similar odors. Unsupervised techniques are not popular because they have less pronounced practices. Interestingly, they got contention about cyber security.
What is reinforcement learning?
Finally, we have reinforcement learning, which is the final limit of machine learning. A reinforcement algorithm is learned through trial and error to achieve a clear goal. He tries many different things and is rewarded or punished depending on whether his behavior helps him achieve his goal. This is like treating a dog when teaching a new trick and withholding. Reinforcement learning is the foundation of Google’s AlphaGo, a program that famously defeats the best human actors in Go’s complex game.