In this article, we’ll explore Convolutional Neural Networks (CNNs) and, at a high level, look at how they are driven by the structure of the brain. If you want to read more about the brain in particular, there are more resources to help you at the end of the article.
We are constantly analyzing the world around us. Without conscious effort, we make predictions about what we see, and act on them. When we see something, we label each object based on what we have learned in the past. To illustrate this, take a look at this picture for a moment.
You’ve probably thought something like “He’s a happy little boy standing on a chair”. Or maybe you thought he was screaming, about to attack this cake in front of him.
This is what we do subconsciously throughout the day. We observe, label, make predictions and recognize patterns. But how do we do that? How can we explain what we see?
It took nature over 500 million years to create a system to do this. The collaboration between the eyes and the brain, called the primary visual pathway, is the reason why we can perceive the world around us.
While vision begins in the eyes, the actual interpretation of what we see occurs in the primary visual cortex in the brain.
When you look at an object, light receptors in your eye send signals via the optic nerve to the primary visual cortex, where the input is being processed. The primary visual cortex senses what the eye sees.
It all seems so natural to us. We hardly think about how special it is that we get to recognize all the things and people we see in our lives. The deeply complex hierarchical structure of neurons and connections in the brain plays a major role in this process of remembering and labeling objects.
Think about how we learned, for example, what an umbrella is. Or a duck, lamp, candle, or book. In the beginning our parents or family told us the names of objects in our immediate environment. We learned from the examples given to us. Slowly but surely we began to recognize certain things in our environment more and more often. They became so common that the next time we saw them, we immediately knew what the object was called. They became part of our model on the world.
convolutional neural network
Just as a child learns to recognize objects, we need to show an algorithm to millions of pictures before we are able to generalize the input and make predictions for images it has never seen before.
Computers ‘see’ differently from us. In their world there are only numbers. Each image can be represented as 2-dimensional arrays of numbers, known as pixels.
But the fact that they see images in a different way doesn’t mean we can’t train them to recognize patterns like we do. We just have to think about what an image is in a different way.
To teach how to recognize objects in images, we use a specific type of artificial neural network: a convolutional neural network (CNN). His name refers to one of the most important tasks in the network: determination.
Sensory neural networks are inspired by the brain. In the 1950s and 1960s on the brain of mammals, D.H. Hubel and T.N. The research by Wiesel suggested a new model for how mammals perceive the world visually. They showed that the visual cortex of the cat and monkey is comprised of neurons that respond specifically to neurons in their direct environment.
In their paper, they describe two basic types of visual neuron cells in the brain that each function in a different way: simple cells (S cells) and complex cells (C cells).
Simple cells are active, for example, when they recognize basic shapes as lines in a certain area and at a specific angle. Complex cells have large receptive fields and their output is not sensitive to the specific position in the region.
Complex cells continue to respond to a certain stimulus even when its absolute position on the retina is changed. Complex refers to the more flexible in this case.
In vision, a receptive field of a single sensory neuron is the specific area of the retina in which something will affect the firing of that neuron (that is, activate the neuron). Each sensory neuron cell has identical receptive fields, and their areas are overlapping.
In addition, the concept of hierarchy plays an important role in the brain. Information is stored in hierarchical order, in the order of patterns. The neocortex, which is the outermost layer of the brain, stores information hierarchically. It is stored in similarly organized groups of neurons in the cortical column, or neocortex.
In 1980, a researcher named Fukushima proposed a hierarchical neural network model.