AI, Artificial Intelligence, Machine Learning, ML, Neural Networks

A Primer Into Neural Networks


Our previous articles have examined what Artificial Intelligence (AI) and Machine Learning (ML) are and how they operate. In this article, we take a step into another direction of AI, which is known as Neural Networks, or NNs for short.

What Is A Neural Network?

In very broad terms, a Neural Network (NN) can be defined as follows:

“They are computing systems with interconnected nodes that work much like neurons in the human brain. Using algorithms, they can recognize hidden patterns and correlations in raw data, cluster and classify it, and – over time – continuously learn and improve.”

(SOURCE:  1)

The Neuron

At the core of any NN is what is known as the “Neuron.” These can be viewed as the underlying foundation of the human brain and the Central Nervous System (CNS), which includes the spinal cord. Although it is difficult to get an exact number, neuroscientists have estimated that there are well over 100 billion Neurons that are in existence in any individual.

The basic objective of these cells is to receive all of the inputs from the external environment and convert these into various types of electrical signals that can be processed by the human brain for our thought and reasoning processes. The Neurons are also responsible for sending commands and instructions to our body muscles so that they can move in the appropriate fashion.

It’s important to keep in mind that the development of Neurons is an ongoing process and is known as “Neurogenesis.” These cells do not remain in a static state as they are always continuing to evolve even well into the late years of adulthood.

The Components of a Neuron

The Neuron is made of three key components:

1. The Dendrites:

This is where the Neuron receives its input from the outside world.

2. The Axon/Soma:

This is where the output of one Neuron is delivered to the next Neuron.  In this regard, these cells don’t operate on their own, they are very interdependent amongst one another. The way that this communication process occurs is known as the “Action Potential”, where the messages are merely electrical charges.

3. The Spines:

The Dendrite also extends out, like branches from a tree. These are known as “Spines” and their primary role is to facilitate the communications process of the Neurons.

The Different Kinds of Neural Networks

The primary goal of any NN system is to replicate the Neuron as much as possible, even including the three components just described. Like Machine Learning (ML), an objective of NNs is to try to make some prediction (which is known as the “Output”) based upon the input that it is being fed.  In this regard, there are three main types of NN systems, which are as follows:

1. The Feed Forward Neural Network (FNN):

This is often deemed to be the most basic type of NN. In this model, the information and data that is collected from the outside world (which are typically large-scale datasets) travels in only one direction – from the input to the output.  There could also be hidden layers in between as well.  

2. The Recurring Neural Network (RNN):

This is considered to be a step up from the FNN.  With this system, the same process is repeated in an iterative fashion, but the key difference here is that the output is not dependent strictly on the input that is being fed into it. Rather, it is heavily dependent upon the reliability of the former processes that have just transpired.

This kind of NN is used quite a bit today when it comes to Natural Language Processing in Chatbots.

3. The Convolutional Neural Network (CNN):

Convolutional neural networks are deep learning networks that are typically used for image processing. They have three main types of layers, which are:

  • Convolutional layer
  • Pooling layer
  • Fully Connected layer

The complexity of the network increases with each layer, with each being able to identify specific portions of the image. 

How A Neural Network System Works

It’s important to keep in mind that NN systems can be very complex with the main factor being the application that it is serving. But to illustrate how it works, we can consider a very simple example. Suppose you are a financial trader and want to predict as much as possible future prices. The input will consist of prices of a wide breadth of stocks, across many industries.  

At the Hidden Layer stage, different correlations and trends will be discovered and sent forward. 

The output will be the anticipated price(s) that have been computed based upon the kinds of permutations, or variables that have been set forth. This is typically done using the principles of high-level statistics, such as Monte Carlo and Bayesian techniques.  

One of the other goals of an NN system is to keep learning on a real-time basis, so it does not go “stale”. Using our example, it needs to be fed massive amounts of pricing data daily, so that it can adapt to different financial market conditions to yield an accurate, predicted price level.  

Other Examples

Extrapolating the above example on a much grander level, NN system can be used to solve the unique needs of the following applications:

  • Fraud that takes place in the medical insurance industry.
  • Determining the most appropriate and direct transportation routes for logistics and supply chain carriers.
  • Predicting what energy demand could be in the future in order to set reflective pricing targets.
  • Quality control in the manufacturing processes.
  • Determining the best marketing approach to take based upon previous feedback from both customers and prospects.
  • Treating cancer patients by predicting how severe the ailment can get.

What to take away

We broke down in detail just how a neural network system works. We also examined the role its core, the Neuron, and its three key components play. We also looked at the different kinds of neural network systems and how they function.

Neural networks are a key component of AI, in particular, machine and deep learning, as they reflect how the human brain operates, providing computer programs with the ability to solve common problems and recognize patterns.

These artificial networks can learn events and make decisions by learning from commenting on related events. Neural networks possess numerical strength with the capabilities to perform multiple jobs and tasks, making them an attractive component of AI.

Stay tuned for our next article will examine yet another area of artificial intelligence – that of Computer Vision.  



Anthony Figueroa

Want to work with Anthony Figueroa? We are hiring :)

Other articles you might like