A New Neuron Model Based on Dendritic Mechanism and Its Applications
Sha Zi-Jun · Institutional Repositories DataBase (IRDB) · 2015
All along, many scientists, one after another, in order to clarify the functions of the human brain to make a lot of effort.From the appearance of neural networks to the study of brain waves, the exploration of the brain has not stopped.With the development of science and technology, more and more new technologies are being used on the research of brain.Research in this area has attracted more and more attention.In order to understand the process of brain, the principle of the complex neural systems, many researchers will shift attention to the neurons, the basic building blocks of the nervous system.Breast cancer is a preponderant disease in the world, and it is one of the most major death causes of women.In order to predict the cancer in women, recent years, artificial intelligent (AI) has been widely used in the scopes.This thesis deals with the application of a novel a neuron model based on dendritic mechanism for classifying breast cancer on Wisconsin breast cancer database (WBCD).As its name suggests, the dendrites mechanism is the main computation of the neuron model.The model neuron is composed of a set of independent branches and a soma.Instead of being weighted simply, the inputs of the neuron model are processed nonlinearly rather than being weighted simply to realize excitatory synapse, inhibitory synapse, constant-1 synapse or constant-0 synapse.The signal of each branch is weighted and performed due to the input.The Soma receives the signals transmitted from the branches to produce the output.The performance of the neuron model based on dendritic mechanism is compared with the classic back propagation neural networks (BPNNs).Simulation results indicate that the neuron model based on dendritic mechanism holding superior capability at the accuracy, convergence speed, stability and AUC.In addition, it is worth of note, through learning, an arbitrarily dendrite of neurons with different initial synapses can develop an internal structure which depends on the location of synapses in the branch, and the type of synapse.Furthermore, in this simulation, the developed structure may suggest some inspirations to the detection of breast cancer.In addition, owing to non-decrease on classification accuracy after ３ eliminating the useless branches of the neuron model, the computation load can be released.In this thesis, the trading data from January 2004 to October 2014 in the Shanghai stock market is selected to verify the overreaction on Shanghai stock market.And, the overreaction from 2007 is found to turn to weaken with the time going by and the influence of the overreaction turn to disappear from 2011.Moreover, the neuron model based on dendritic mechanism, for the first time, is also proposed to fit and predict the changes about abnormal returns of ill-performed and well-performed stocks in test period.The result shows that the neuron model possesses high computational ability and successes to predict the tendency of overreaction.