Transienti in Reti Neuronali Basate su Attrattori

The ability to process sequential information has always been considered as one of the most important features of biological and artificial intelligence systems. Although the investigation of learning and memory has a long history, little is known about possible dynamic principles of learning and remembering multiple events and their time sequence by neuronal systems. Here we propose a model based on results from the theory of nonlinear dynamics in neuroscience, in particular the concept of Winnerless Competition (WLC), which is an adequate candidate for sequential spatial memory. The essence of this idea is that sequential memory is encoded in a high-dimensional dynamical system which shows complex heteroclinic orbits that connect a sequence of saddle points. Each saddle point represents an event or pattern to be remembered. The existence and global stability of such a heteroclinic skeleton can be proven given the presence of asymmetric inhibitory connections. These connections are inevitably shaped by sensory inputs which sense the environment. In conclusion we will further support the plausibility of this model by showing its multifractality properties and how those can be connected to optimal search strategies.