Brains vs. Bytes: The Tale of Neurons and Networks

The origin of artificial neural networks was the algorithms that try to mimic the brain. So, scientists started to build a structure that is similar to the brain neurons' activities based on their knowledge at that time.  

In the brain structure, there are trillions of neurons that are connected. Each neuron has multiple input tentacles, which are called dendrites, multiple output tentacles, which are called axons, and a central part for creating the biochemical reaction based on input signals and passing it through outputs. Artificial neural networks were created based on this structure. They contain multiple inputs and outputs. There is a weight for each input to control the effect of input neurons, which is a similar process that happens in the brain for input signals. Finally, there is a section that performs a nonlinear function on the weight and the input neuron and passes the results to the output.  

As you see in the above image, artificial neural networks and biological networks are alike in many aspects. But they also have several differences.

Difference between Biological Neurons and Artificial Neurons

Although Artificial neurons were inspired by biological processes, they are different from their biological counterparts in several ways. The brain usually contains trillions of neurons, but the artificial neural network has much less than this number. Furthermore, the structure of artificial neurons is a way that there isn't any guarantee that increasing the number of neurons can increase the performance. They also have different topologies, power consumption, learning policy, etc. But one of the other most critical differences between brain neurons and artificial neurons is the ability of self-regeneration in the brain. The brain uses a similar architecture for managing different tasks, which means the part of the brain that controls the sense of taste can also manage eyesight if we give it suitable signals as input.  

Moreover, this characteristic helps the brain to be robust to false tolerance. For instance, minor failures in the brain will not result in memory loss. On the other hand, the architecture of artificial neural networks is not designed to handle this type of problem. Because, first of all, there are different architectures that work better for different types of data. Besides, artificial neural networks are too sensitive to failure. A minor failure on a layer of an artificial neural network can change the result in a way that becomes useless.

For deeper insights on neurons and networks, make sure to be a part of our Fundamentals of Deep Learning Online Training.