Computer sciences have benefited from the achievements of other fields repeatedly. In 1943, Warren McCulloch and Walter Pitts created the mathematical model of neuron called the Threshold Logic Unit. Since then, many researchers have tried to reproduce the way the human brain works using the discoveries of neurobiology, physics, and mathematics. An appropriate, artificial simulation of the natural nervous system should be able to make it behave similarly to the human brain and perform sophisticated tasks.


Basics of Artificial Neural Networks

The structure of neural networks is modeled on the human nervous system and the brain. Its action is based on relatively simple functional units – neurons. The key to simulate the biological structure is to recreate artificial neurons and connect them to work as a whole.

There are about 10 billion interconnected neurons in human brain. Each neuron is a cell that uses biochemical reactions to receive electrical signals through dendrites and transfer them through the long thread-like parts called axons. Each axon is connected to other neurons across a small gap – the synapse, which converts the activity from the axon into the electrical charge that inhibits or excite activity in the connected neurons.


Artificial neural networks are loosely based on this biological structure. They consist of some interconnected nodes (artificial neurons) arranged in several layers. Each neuron within the network is a simple computing unit which takes one or more inputs and produces an output. At each neuron, every input has an associated weight which modifies the strength of each input. The neuron adds together all the inputs and calculates an output to be passed on.

Computers and Artificial Neural Networks – Differences

The main difference between conventional computers and ANNs lies in the process of learning. While standard systems require programming, the neural networks can learn from examples. Because the neural computers can adapt themselves during the training, it does not require the problem to be described explicitly.

ANNs operate by inductive reasoning – rules can be constructed based on given training examples. Computational processes are asynchronous and parallel. Because of the shared responsibilities, it is also more fault-tolerant, while in conventional systems single error can affect the work of the entire program. All the above features make ANNs able to handle noisy or incomplete data.


Possible Applications of Neural Networks

Artificial neural networks can be applied for many tasks that require pattern recognition, classification, or forecasting. Practical applications of ANNs vary from commercial use in marketing (consumers habits, forecasting sales), financial markets (stock market predictions), medicine (diagnosis, interpreting ultrasound or electrocardiogram images), or even military (motion detection, radar image processing).


In the recent years, Google has been strongly interested in the development of neural networks. In 2016, they introduced Neural Machine Translation System, which significantly reduced the number of machine translation errors in Google Translate service. Neural networks are also applied in handwriting and voice recognition in cell phones or anti-spam systems.

The wide array of possible applications of neural networks makes this technology one of the main priorities for many leading IT companies. All in all, who would not want to make computers a little bit more human?


This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.