In this thesis, the continuous attractor neural network (CANN) model is applied to investigate neural field dynamics in the presence of short-term synaptic depression (STD). There are three parts related to neural field dynamics. They are: 1) intrinsic dynamics, 2) response to a single static input, and 3) response to two overlapping inputs. In the first part, intrinsic dynamics in both strong global inhibition and weak inhibition scenarios are explored. Different dynamics of these two scenarios are related to the ratio of the neuronal time constant to the STD time constant. In the second part, response to one single static input is investigated. Four basic kinds of response patterns and more complex mixtures of them are discovered. By monitoring the period of the response dynam...[ Read more ]

In this thesis, the continuous attractor neural network (CANN) model is applied to investigate neural field dynamics in the presence of short-term synaptic depression (STD). There are three parts related to neural field dynamics. They are: 1) intrinsic dynamics, 2) response to a single static input, and 3) response to two overlapping inputs. In the first part, intrinsic dynamics in both strong global inhibition and weak inhibition scenarios are explored. Different dynamics of these two scenarios are related to the ratio of the neuronal time constant to the STD time constant. In the second part, response to one single static input is investigated. Four basic kinds of response patterns and more complex mixtures of them are discovered. By monitoring the period of the response dynamics, relations among them in the phase diagram are presented. The effects of STD and the input of different strengths on the response are discussed. In the last part, the ability of these networks to better resolve two overlapping inputs are presented. Resolution enhancement is enabled by population spikes, one of the four basic response patterns to one single static input. When two overlapping inputs are imposed, population spikes will switch back and forth between the two inputs, thus facilitating temporal coding, which preserves more information than traditional time-averaged rate coding. Several conditions for the resolution enhancement in our model are discussed and we argue that resolution enhancement can be achieved in other models as long as those conditions are met.