Real time VS forecast

Real time and forecast may seem very different concepts because they refer to different timeframes. We always think of real time as the present, while a forecast is a prediction about the future. A forecast is only worth something if we can benefit from in the present (real time).

When talking about neural networks, forecast and real time complement each other. In general, when we observe an event, or evolution of a system, for instance the flight of an aircraft, we might find it useful to assume that it does not change its state, or in other words, it does not evolve while we study it.

To build a mathematical model of the system, it is necessary to analyse a series of exact data about the past behaviour of the system, that will support the forecasting process. The neural network is based either on a pre-existing mathematical model, ready to be queried, or on a new model that is updated in real time. Its task is to store some information in memory and process new ones to build a reliable forecasting algorithm. This means that an algorithm, built through a neural network, if properly queried, will provide a valid and useful answer for subsequent processing.

A neural network can process large amount of data, but it cannot understand its meaning. The observer's contribution is therefore essential to allow the neural network to process the right data. A neural network, as computational tool, will keep on forecasting using the data it receives in real time from the environment. Interestingly enough, a human observer will use its biological neural network (nervous system) to do just that!

In conclusion, we can take advantage of the response and real time update capacity of a neural network to create more and more accurate forecasts, just think of Covid-19 infection rate predictions! As mentioned an interesting peculiarity is that the processed data are not strictly dependent on each other, the network does not “understand” any physical relationship between the variables, but tries to find a mathematical relationship between them. The dependence between the input data is made explicit only in the final result. In fact, the network is capable of building linear and non-linear relationships even if it is not able to recognize the real physical principle between the causes, i.e. it cannot understand why a certain event happens. The output of a neural network is not the interpretation of a generic event but a generalization of that event, by taking into account numerous variables, or sensory information. The interpretation of the results is still left to the observer.