The learning elements that constitute the building block of Neural Networks are similar to the brain cells in humans. The similarity, coming from how the brain works as far as learning, memory, recollection, understanding and making associations are concerned. The brain’s primary equipment, is the enormity of cells – brain cells (100 billion cells) – that it has at its disposal. To be precise, the more the amount of cells in an individual’s brain, the more learning they would be able to accomplish, with deeper associations and understanding. The reason for this being that, brain cells get used, when learning takes place and when memory is formed. Such that if the brain, runs out of these cellular structures, assimilation becomes difficult. Under such circumstances, recreation and rejuvenation becomes necessary, if brain work is to continue.
To add to this, thoughts are created when sensory information from the body, arrives at the brain and creates cellular clusters. Which are unique patterns, that become the building block of memory, formed as a cerebral version of the impulse received. After much of this, has been stored, a collection of these become memory, and can be replayed or recalled. This is made possible, based on the nature of used brain cells, which form geometrical arrangements. These when energised by the central nervous system, begin to play back any captured sensory input. They take the form of an emotion or recollective imagination.
This is why addictive substances, such as alcohol, which remove brain cells, are really not good. As they equally remove brain cells, that have become part of a memory, an emotion or a though. Such people complain of memory loss, loss of cognitive ability and detachment from past experiences.
The same science is used in building machine learning algorithms, which employ neural networks, to emulate the human brain and its learning ability. Neural networks, would be able to gather on speed when compared to the brain in performance. The human brain, at the highest level of focus and concentration, operates at 30 Hertz. Where as modern day computer CPUs, which are the housing for neural network algorithms, operate at 2 Giga Herz (2,000,000,000 Hertz) speed. Making the learning of more intricate data possible, using algorithms.
Neural networks, learn by substituting brain cells with numbers in a N by N matrix formation. They recognise patterns in a problem, which is captured as data, and converted to a numerical vector. This data could be in the form of text, sound, an image or time series. Such that, for each feature in the data being learned, a neural network matrix assumes a cluster formation. Which can result in several layers of clustering, for the learning and recognition of more complex and intricate patterns. An approach known as deep learning. This mathematical analogue, that works similar to the human brain, is known as the algorithm.
Algorithms with sufficient learning elements, such as the human brain, can be used to automate anything from labelling email as spam or not spam to the self driving of vehicles. If they are trained with a large enough training set, having a good set of features to learn from. With this approach, algorithms come up classifications such as the RGB values, of pixels, of images and the ID numbers of persons. The predictive analysis of the pattern in a timely series of recurring event, such as customer churn, employee resignations, health issues, machine break downs and dates when they are most likely to re occur. The clustering of products for anomaly detection or people for behavioural traits that could lead to fraud.