The neural networks technology

The theory of neural networks creation started from the idea of designing intelligent computing devices like biological systems (usually it's compared with the human brain because a neural network simulates brain cells inside the computer so that it learns by itself). This theory became the most useful for artificial intelligence development.

Neural network and its components

Neural network consists of neurons (units) distributed among the layers. There are 3 types of units: input, output and hidden ones. Input units receive the information (training data), output units issue the received processed information. An artificial neuron receives signals from many inputs, processes them in a unified manner and transmits the result to many other artificial neurons, i.e. does the same thing as a biological neuron.

These units are connected with the hidden units on the different layers that transfer the information. The connection is represented by numbers called weights. So the unit of the first layer changes the unit of the next transferring the information having its own weight. The input information is processed and turned into a result and the training of the neural network is based on the experimental selection of such a weight coefficient for each synapse which leads to the desired result.

The number of layers illustrates how complex the neural network is. More layers result in more difficult tasks that it can deal with. This kind of neural network is able to learn deeply. Using the deep learning neural network Google improved the possibilities and quality of its Translator.

Common tasks for neural networks

Neural networks are already at the very heart of many day-to-day technologies such as automatic license plate recognition systems or handwritten postcode reading systems.

The most common tasks for neural networks are:

  • Pattern recognition – it is used in Google for example when looking for a photo, or in the smartphone camera when it detects the position of a face and highlights it.

  • Classification – the distribution of data by parameters (it is used in a bank system for analyzing information such as age, ability to pay, credit history, etc.).

  • Decision making and management – it's similar to classification, the characteristics are received in input units.

  • Clustering – it's dividing the set of input signals into classes, while neither the number nor the characteristics of the classes are known in advance. After training, such a network is able to determine which class the input signal belongs to.

  • Forecasting – the ability of a neural network to predict and highlight hidden dependencies between input and output data.

  • Approximation– the neural network is able to approximate any continuous function with some predetermined accuracy.

  • Data compression and associative memory– the ability of neural networks to identify relationships between various parameters to present data more compactly.

Neural network to identify tiger mosquitoes

A study by researchers in the Scene Understanding and Artificial Intelligence (SUNAI) research group, of the Universitat Oberta de Catalunya's (UOC) Faculty of Computer Science, Multimedia and Telecommunications, has developed a method that can learn to identify mosquitoes using a large number of images that volunteers took using mobile phones and uploaded to the Mosquito Alert platform this year.

The Mosquito Alert platform can identify the characteristics of a special type of mosquito using photos downloaded on this platform. "The neural network we have developed can perform as well or nearly as well as a human expert and the algorithm is sufficiently powerful to process massive amounts of images," says Adhane.