Neural Networks — Brains like computers, computers like brains.

Neural networks have existed since 1943, however their growth has been quite accelerated since 2006.

These neural networks are the basis for Deep Learning, which in turn enhances the use of Machine Learning.

Neural networks are based on the structure of the human brain to perform its functions.

Although they are not exactly the same, the strategy of use is based on the same structure of neurons that make up a human mind.

Simplified, what is a neural network?

In a very simple way, neural networks receive data, train themselves to recognize the patterns of this information, and predict the outcome for a new set of similar data.

How do neural networks work?

A neural network works on the basis of neurons, which perform the central processing function within the system.

This in turn is divided into layers, where is the input layer that receives the information and the output layer that predicts the final output of the process.

Between these two layers are the hidden layers, which are responsible for carrying out the process of computing the data to transform the input data into a prediction as accurate as possible.

Each layer is connected to each other through channels that assign a numerical value to the relationship known as weight.

This is where it gets a little complicated.

Each of the input data depending on the neuron, is multiplied by the assigned weight and then added depending on the channel junction.

Once this operation has been performed, the result is assigned a number value known as “Bias”.

The result of this process must be analyzed by what is known as the activation function that discriminates which neuron in the layer must be “activated”.

Only the neurons that have been activated are those that transmit information to the neurons of the next layer following the exact same process, only now narrower. This process is known as forward propagation.

This process is repeated until reaching the output layer, where, based on a probability index, the neuron with the highest or most probable value is the one that determines the most accurate prediction that the model will return.

Neural network model training

Despite the fact that neural networks have an efficient shape and a high degree of confidence, their results are not always correct, so they must be “trained”.

Based on the results obtained, whether they are correct or erroneous, these must be measured in order to find a difference index between each of the results.

This information is sent to the first layer of the neural network through a process known as backward propagation.

Based on this indicator, it is possible to constantly adjust the indexes, weights and results of the neural network.

The forward propagation and backward propagation processes are done iteratively, that is, constantly with different types of input data.

In order that in the end, the predicted result is correct as many times as possible.

How long does a training process take?

One of the most complicated processes within the training process is being able to find the adequate data to feed the neural network.

Data cleaning and control of your processes takes the longest.

Once the training process has started, it can take hours, days or even months depending on the degree of complexity and variables to consider.

However, it is important to mention that if the investment in programming and training of a neural network is made, the result can be of great benefit in the long term, despite the long time it may take to create it.

It is important to mention that the training process must be a cycle, since the data can become obsolete over time.

Applications of a neural network?

• Facial Recognition
• Prediction (Climate, Financial, etc.)
• Market and customer profiles
• Predictive maintenance in the industry
• Scientific research.

A bit of science fiction

As mentioned at the beginning, the objective of a neural network today is to be able to replicate in an optimized or similar way the functioning of the human brain.

This type of progress suggests that at some point you might think that the human brain can be replicated or “emptied” into a neural network in order to preserve the human mind in a non-mortal entity.

There is great expectation within the scientific community to be able to achieve this, we have even seen it already in movies like Transcend or series like Altered Carbon.

There are several questions to consider, such as consciousness, mind or even soul, but the fact of creating an interface that allows a human brain to be duplicated in a synthetic brain, is undoubtedly a step that is sought to be taken in order to be able to:

• Achieve eternal life .
• Know the universe
• Preserve the human race.