1.8. How much artificial neural networks are simplified in comparison to biological ones? (Translation by Ryszard Tadeusiewicz; rtad@agh.edu.pl ) Neural networks are of f course very simplified in comparison to real neural systems of most living creatures. It can be observed on Fig. 1.28, where region occupied by typical neural networks (realized as usually as programs for general purpose computers) marked as yellow square in coordinates showing on abscissa structural complexity of the considered neuroinformatic system and on ordinate speed of the system functioning. Both dimension on this plot are represented in logarithmic scale because of huge distance between smallest and biggest presented values. For example structural complexity measured by number of synapses in considered neuroinformatic system can vary from 102 for typical artificial neural network used for technological purposes up to 1012 for the human brain. This dimension for artificial neural networks is limited by value about 10 5 – 106 because of computer memory limitations, where appropriate values for not very complicated “brains” of fly or bee can be characterized by numbers of synapses 108 – 109 respectively. In comparison with these neural systems brains of mammals are really huge with 1011 synapses for rat and 1012 synapses for human central nervous system. Let consider almost linear relation between structural complexity of such (taken into account) biological neural systems and speed of the their functioning (Fig. 1.28). In fact it is general rule, caused by massively parallel method of biological neural systems functioning. For this type functioning when system consists of more elements (more neuron and more synapses) and all these elements working together (simultaneously) – speed of data processing increases proportionally to the system structural dimension. Fig. 1.28. Localization of artificial neural networks and selected real neural systems on diagram showing relation between Structural complexity of the system (number of synapses) and speed of the system functioning For artificial neural networks speed of system functioning depend on the form of network realization. When neural network is realized as a program simulating neural activity, learning, problem solving etc. on general purpose computers (including laptops, tablets and palmtop devices) – the functioning speed is limited by performance of used processor. It is evident, that is impossible to speed up processing time over hardware limitations using any type of programs, therefore artificial neural networks realized as a programs on general purpose computers are rather slow. Is possible achieve very fast functioning of artificial neural networks when there are realized in hardware form (see blue ellipse on Fig. 1.28). In bibliography or in internet you can find many examples of neural networks realized as specialized electronic chips – recently often in FPGA technology. Are known also optoelectronics solutions, chips fabricated using partially analog technologies (most systems are of course digital taking into account input , output and general control of the system , but sometimes smart analog devices inside can be incredible fast). Known are also neurochips made from both electronic silicon part and biological part – living neural call or neural tissue piece treated as a device component. All such methods of hardware realization of artificial neural networks can be very fast (see Fig. 1.28) but the structural complexity of such systems is always very limited and elasticity of system application is also not satisfactory for most users. Therefore in practice almost all users of artificial neural network prefer software solutions accepting their limitations. Looking on Fig. 1.28 we can see, that some of biological neural systems can be located within artificial neural networks range. For example “brain” of the shrimp can be compared with artificial neural network and have no superiority nor in the complexity domain, nor in sense of speed of information processing. Therefore red dot symbolizing shrimps neural system parameters is located inside yellow area symbolizing artificial neural networks parameters. But for most biological species complexity of their neural systems and speed of data processing are much greater than best parameters achieved by artificial neural networks. In case of human brain its complexity is billion times greater, than parameters observed in artificial neural networks. Therefore thinking about natural and artificial neural networks we should be very humbly!