Learning vs. Intelligence
Artificial intelligence is a field of study quite similar to biology or chemistry.
Machine learning is a part of artificial intelligence. An important part, but not the only one.
Neural networks are one type of machine learning. It's popular, but there are other good guys in the class too.
Deep learning is a modern approach to building, training, and using neural networks. Basically, it's a new architecture. In practice today, nobody separates deep learning from "regular" networks. To avoid sounding like a dumbass, it's better to just name the type of network and avoid buzzwords.
|
What a machine can do |
What a machine can't do |
|---|---|
|
Prediction |
Create something new |
|
Memorization |
Become intelligent very quickly |
|
Reproduction |
Go beyond their task |
|
Choose the best option |
Kill all humans |
Map of the Machine Learning World
If you are too lazy to read for a long time, take a look at the image below to get a certain understanding.
Remember, it's always important - there is never only one solution to a problem in the world of machine learning. There are always several algorithms that are suitable, and you need to choose which one is more suitable. Certainly, everything is solvable with a neural network, but who will pay for all those GeForces?
-
Classic Machine Learning
The first methods came from pure statistics in the 50s. They solved formal mathematical tasks - searching for patterns in numbers, evaluating the proximity of data points, and calculating vector directions.
Today, half of the internet works on these algorithms. When you see a list of articles to "read next," or your bank blocks your card at a random gas station in the middle of nowhere, it's most likely one of these guys at work.
Large tech companies are huge fans of neural networks. Specifically. For them, 2% accuracy is an extra 2 billion in revenue. But when you're small, it doesn't make sense. I've heard stories about teams spending a year on a new recommendation algorithm for their e-commerce website before realizing 99% of the traffic was from search engines. Their algorithms were useless. Most users didn't even open the main page.
Despite their popularity, classic approaches are so natural that you can easily explain them to a toddler. They're like basic arithmetic – we use it every day without even thinking about it.
Supervised Learning
Classic machine learning is often divided into two categories - supervised and unsupervised learning.
In the first case, the machine has a "supervisor" or "teacher" who gives all the answers to the machine, like whether it's a cat or a dog in the image. The teacher has already divided the data into cats and dogs (labeled), and the machine uses these examples to learn. One by one. Dog by cat.
Unsupervised learning means the machine is left alone with a pile of animal photos and a task to figure out who's who. Unlabeled, no teacher, the machine tries to find a pattern on its own. We'll talk about these methods below.
It's clear that the machine learns faster with a teacher, so it's used more in real-world tasks. There are two types of such tasks: classification – predicting the category of an object, and regression – predicting a specific point on a numerical axis.
