Research On Modular Latching Dynamics Model In The Cortical Networks

Inspiration always stems from nature. The development history of artifcialintelligence just witnesses the process of human learning from the natural intelli-gence. And it seems such a kind of pursuit will never stop in the future. In the longriver of human evolution, the cortex grows thicker, and further into lamination. Theincreasing speed of cortical surface and volume have being accelerated in the pastten thousands of years. About50thousands years ago, the humankind departedfrom their brothers, apes. Though there are no qualitative diferent traits that candiferentiate humankind from their clan relatives, but the language ability, whichis considered to be the most outstanding intelligence in the nature, is amazinglyunique for the humankind. Human language is able to construct infnite sentencesfrom fnite gradients. And how to model such an ability is the frst task towards thefuture studies on languages.Latching chain, designed by Alessandro Treves from SISSA using Potts net-work, is a kind of adaptive dynamics model that derived from the idea of infniterecursion of language. Latching dynamics generates infnite patterns sequences fromfnite pattern set when some specifc network connection conditions are fulflled.Though latching chain well explains the strategy transition phenomenon in actions,thoughts and language, they also face many basic challenges. For example, latchingchain is built on a single network, which is contrary to the imaging evidences onfunctional area divisions. How to construct a more feasible latching dynamics modelaccording to the cortical network connections? In this thesis, a modular latchingmodel is successfully constructed. And the impacts of diferent network structuresto the modular latching chain are also studied. As a support for the modular latch-ing dynamics, the functional networks of auditory modality and the fast-rewiringmechanism underlying the object-cognition are also investigated by analyzing theopen fMRI dataset. The details are described as follows:Firstly, as an efort to help others understand the Potts network, the possibleelectrophysiological mechanism of the discrete states of Potts unit is discussed bynetwork simulation, where the neuron characters and network structure are synthe-sized and constructed from the minicolumn. Simulation results using Izhikevich neuronal model shows that the Potts state may stem from the synchronous spikingof neuronal subsets.Secondly, before the modular extension of Latching dynamics, the function-al network structure of auditory modality is studied by analyzing the open fMRIdataset. One-step mean-shift clustering algorithm is proposed to cluster the fMRItemporal signal. Results show that the functional subnetworks in both the basic-level activation and the subordinate-level activation follow the modular scale-freedistribution.Then it comes the central part of the thesis, the modular extension of latchingdynamics. Taking the modular small-world network structure into consideration,we successfully design the weight formula of the hetero-associative connection. Theintroduction of transmission delay and adaptive threshold as a global feedback makesthe system more similar to the actual neuronal information processing mechanism.As a carrier supporting the modular latching dynamics, the dynamics features ofcerebrum (M-network) are studied by computation simulations.After the introduction of modular latching chain, the dynamics diferences be-tween cerebrum and cerebellum (K-network) are compared. Simulation studies showthat the latching chains in the K-network have relative more fxed transition path-way, while they are more fexible in the cerebrum. The length of latching chainsof M-network is robust to the changes of rewiring probability. On contrary, on-ly a small range of rewiring probability supports longer latching chains. Besidesthat, M-network is more robust to the noise pattern pairs and feedback connectionsthant K-nework. These result hints that the specifc connection structure maybeone important factor leading to the advent of human cognition.Lastly, we discuss the probable switching mechanism underlying the fast cogni-tion to the complex stimulus. There would be directed and fast latching-switchingwhen outer stimuli enters. So what mechanism support the fast-rewiring of func-tional network in the brain? Through the open fMRI dataset analysis, we concludethat there may be hierarchical activation mechanism. That is, fast-rewiring maydepends on some skeleton voxels.