So what is a Neural Network anyway?

AI is everywhere nowadays, but have you ever wondered how it actually works? It turns out that, much like our brains, AI works using something called a “neural network”. Let’s pop the AI hood and take a look inside to figure out how it works!

What is a Neural Network?

First off, let’s talk about what a neural network even is. Basically, it’s a type of AI that is designed to work like the human brain. Just like our brains have neurons that help us think and learn, neural networks have artificial neurons that help AI learn and make decisions. These artificial neurons are organized into layers, and each layer has a different job.

So How Does a Neural Network Work?

Just like the brain in your head, a neural network has different layers. Each layer is made up of nodes, which are like mini-models themselves! These nodes are all connected to each other, and they help the neural network process information and learn from it.

Here are the main parts of a neural network:

1. The Input Layer: This is where the neural network receives information. For example, if you want the AI to spit out some great copy, you tell this layer what you want written.
2. The Hidden Layers: These are the layers between the input and output layers. They do the hard work of figuring out what the information means. Each hidden layer is made up of nodes, and each node processes a piece of the information. There can be many hidden layers, depending on how complex the task is. AI with only one hidden layer is called “shallow AI”, while AI with more than one hidden layer is called “deep AI”.
3. The Output Layer: This is where the neural network gives its answers, or “output.” Using the copy example, the output layer is the written text that the AI spits out!

Nodes: The Building Blocks of Neural Networks

This explanation definitely raises even more questions. For instance, what’s a node? Well, to understand neural networks, it’s important to understand what a “node” is.

Each node, represented by a colored ball in the illustration above, is like its own tiny computer. It takes in information, processes it, and then passes it on to the next node. One common way nodes process information is by using something called linear regression. This is a term for taking information, such as from a dataset, and using it to predict what will happen next.

Weighted Nodes

Not all nodes are created equal. Some nodes are more important than others, and this importance is called their “weight.” A node with a higher weight has more influence on the final decision of the neural network. 

To understand this, let’s use an example (caution - math incoming!). I promise it’ll be simple!

Let’s say you want to go on a hike over the weekend. You’ve found a trail you’re considering. Do you go?

Well, let’s say you have 3 factors that you need to consider before going. Each factor is worth 1 point for “yes”, and 0 points for “no”. 

• Is the trail at your comfort level? Let’s assume it is, so it gets a 1.
• Is it close enough that you’ll be able to complete the trail and be home in time to go to your friend’s birthday party? Let’s assume it is, so that gets a 1 as well.
• Finally, are there bears? There haven’t been any reports of bears, so that’s a 0.

Now, you view these factors are all of different levels of importance to you. Let’s say that, on a scale of 1-5, the comfort level is a 2, being at your friend’s birthday party is a 4, and the bears are a 3. Now, all of these factors have a “weight” associated  with them!

Now, here comes the math part. To determine whether or not to go on the trail (a yes/no, or “binary” question), we are looking to figure out if all of these numbers and their weights add up to more or less than 1. If it’s more than 1, you’re going. If it’s 0 or below, you’re staying home!

So, we have the trail comfort level (1X2) plus the ability to go to your friend’s birthday party (1X4) plus the fact there aren’t any bears (0X3). You now subtract 3 because that’s how many factors there are. The formula looks like this.

(1X2)+(1X4)+(0X3)-3 = 3

All the factors and their associated weights equal 3, which is more than 0, so it looks like you’re going on that hike!

Pretty simple right?

Well, this is how a neural network works - just with a TON more factors and a whole bunch of different weights.

Different Types of Neural Networks

There are lots of different types of neural networks, each designed for specific tasks. Three of the more common ones are: 

1. Artificial Neural Network (ANN): This is the basic type of neural network. They are good at general tasks like recognizing patterns in data.
2. Convolutional Neural Network (CNN): This is used mostly for image recognition. They’re great at looking at pictures and figuring out what’s in them. If you’ve ever used a phone app that can tell what’s in a photo, it’s probably using a CNN.
3. Recurrent Neural Network (RNN): This is used for tasks that involve sequences of information, like understanding sentences or predicting stock prices. It remembers what it’s learned before and uses that memory to make better predictions. This is how Propel is able to generate pitch and press release drafts!

Neural networks are a powerful tool in the world of AI. They help computers learn from data and make smart decisions, just like our brains do. By understanding the basics of how neural networks work, we can better appreciate the amazing technology that powers many of the devices and applications we use every day. Whether it's recognizing a friend's face in a photo or generating pitches and press releases, neural networks are at the heart of it all.