New Neural Nets

Abstract

Conventional neural networks work by changing the synaptical weights between their neurons. New neural nets (NNN) are presented, using the recording of temporal sequences of activity, generated by various patterns in chains of neurons, to store and reproduce those patterns. In a biological hypothesis, recording of temporal sequences of activity could be done by nucleic acids, each base triplet encoding a certain degree of activity. The specific properties of nucleic acids would enable those systems to perform an associative data processing. Hybridization-processes will select memory-strings with similar sequences, using the high affinity between homologous complementary strings. Using selected homologous memory-strings, the neural chain is able to give an associative “answer” to a presented pattern. Many neural chains, working together, will form a NNN. To demonstrate their potential, the results of a preliminary study are shown, concerning the occurence and treatment of aphasic symptoms in NNN after a network-lesion.