Originally published on
If you dip even a toe into the realm of artificial intelligence, you’ll come across artificial neural networks.
Artificial neural networks are the systems that power artificial intelligence. It’s a type of computer that doesn’t just read code that it already understands. Neural networks process vast amounts of information to help create an understanding of what’s already right in front of you.
People think the key to understanding neural networks is calculus, but this system of computing has roots in biology. The human brain inspired scientists who created these systems.
So, what better way to explore computing than by thinking about the way we process information?
What Are Artificial Neural Networks?
An artificial neural network is a type of computing system. It’s made up of simple processing elements. These elements are also highly interconnected.
Interconnectivity lies at the heart of neural networks because it helps the system operate dynamically. The dynamic nature of the system also makes these machines challenging to comprehend. Fortunately, there’s a simple comparison available.
Scientists involved in designing the original networks were inspired by nature. Their inspiration comes from the human brain in particular. Our brains host billions (at least) of neurons that connect with every single cell in our bodies. Those neurons also communicate with each other in a dynamic way that allows our bodies to function.
Artificial neural networks operate on a far simpler plane compared to the human brain. A system will only include thousands or hundreds of thousands of connections compared to the billions found in the human brain.
Moreover, the processing power is presently far weaker than your cerebral processing power.
How Neural Networks Work: Layers Of Learning
Neural networks feature layers and layers of connections. Each layer features an input layer to receive information and output layers to share the information. Hidden layers process all the information in between.
Input nodes receive patterns from elsewhere. The input nodes then use connections to the hidden layers to process, decode and reassemble the information. Information isn’t just sent onto the closest nodes. Nodes are weighted to send recognizable data to the right places.
Here’s an example:
Walking down the street, you look down and see a golden retriever. How do you know it’s a golden retriever? Well, you were taught to use logic to recognize things. The “data” presented to you shows that it has four legs. It also has a furry body and a long, feathered tail. The cheerful face is also consistent with what you know to be a golden retriever.
Plus, someone at one point taught you the words “golden retriever,” so you have a name for what you’re looking at.
In the end, your brain processes all these details using logic and comes up with an answer. If it looks like a golden retriever, and it walks like a golden retriever, then it must be a golden retriever.
Artificial neural networks work similarly. They receive information, like “four legs,” and rely on reason based on the information given to establish what’s called a “learning rule.”
Just as you had to be taught that a big, yellow dog with a happy smile is a “golden retriever,” so do networks. A neural network’s learning depends on the type of learning it engages in.
How Is This Different From A Normal Computer?
You might be thinking, “Surely the processing happening here isn’t so far away from the way a conventional computer works.”
The truth is that artificial neural networks are vastly different in both structure and function.
A traditional computer has a central processor that reads instructions given to it from other parts of the machine and the memory. Your computer is merely performing like a trained monkey. If you provide it with information that’s new and unknown, you’ll get an error.
An artificial neural network doesn’t feature a central processor. It uses hundreds of simple ones that collect information from other processors. These networks don’t follow the rules. They respond to the information given to them.
You can see the difference in the dog analogy we just used.
A conventional computer holds a single processor that follows the instructions it already knows. If your laptop saw a golden retriever, it would only know it was a dog because it already knew it was a dog. If your computer never learned about the wonderful world of dogs, it wouldn’t be able to tell you that it was looking at a golden retriever.
Neural networks, however, don’t have to worry so much about it. It receives the number of legs through one input node. Furry faces get processed by another node. The output rationalizes that we’re looking at a golden retriever.
As you can see, neural networks operate closer to the way your brain works compared to a conventional computer. It’s by no means a complete replica, but the inspiration is clear.
How We Use Artificial Neural Networks
Artificial neural networks are interpreters and approximators. They use prescribed learning rules to reason with the data they receive.
Because they’re approximators, their best uses are in interpretations that tolerate errors well. You wouldn’t use a neural network to bake the perfect cake, which requires exact measurements.
Instead, engineers use these networks to discover patterns. These tools are particularly helpful when the amount of data is too significant for humans to see patterns. For example, medical researchers sort through every single published paper about chemotherapy. They can use neural networks to find patterns and associations beyond the subject of chemotherapy.
Artificial neural networks, as they currently stand, don’t create new answers out of existing data. However, they can process data in a way that allows humans to find those answers.
Neural Networks Help Us Learn
Artificial neural networks don’t create new facts. Instead, they help us make sense of what’s already in front of us. Their structure is what enables artificial intelligence, machine learning and supercomputing to flourish.