Discrete and continuous compressed information representation using phase transition phenomena

点击翻译按钮获取中文摘要

A number of findings suggest that the preferences of neighboring neurons in the inferior temporal (IT) cortex of macaque monkeys tend to be similar. However, recently However, Tamura, et al. found that neighboring neurons react to non-similar stimuli lb resolve this controversy, we proposed a new view of information representation in the IT cortex that takes a new view of information representation in the IT cortex that takes sparse and local neuronal excitation into account. Since the excitation is sparse, information regarding visual objects seems to be encoded in a distributed manner. The local excitation of neurons coincides with the classical notion of a column structure. Our model consists of an input layer and a nerve field having afferent connections from the input layer. The afferent connections are learned through Hebbian learning.To check abilities of the present model we employed two parallel rings in the unit sphere as an input space. The input signals have two parameters. … More One is θ; which identifies from what point of the ring the input signal emerged. The second parameter is ξ; which indicates which ring the input signal belongs to. This means that the input space contains the discrete information e and the continuous information B. We define D as distance between the two rings. When the distance D is small, only the position θ on the rings is continuously encoded using the position of local excitation. On the other hand, when the distance D is large, the center of gravity of the local excitation continues to represent the position θ on the rings, and the firing pattern within the local excitation represents ξ in a distributed manner which ring the input signal comes from. It means that there is phase transition regarding the information representationIn order to explore this phase transition phenomenon, we investigated neural network models with mexican-hat and spin glass type interaction using the statistical-mechanical methods, especially the replica method. Here we study the influence of disorder in the mexican-hat type interactions on the cooperative and uncooperative behavior of recurrent neural networks by using the replica method. Although the interaction between neurons has a dependency on distance, our model can be solved analytically. Bifurcation analysis identifies the boundaries between paramagnetic, ferromagnetic, spin-glass, and localized phases. In the localized phase, the network shows a bump like activity, which is often used as a model of spatial working memory or columnar activity in the visual cortex. Simulation results show that disordered interactions can stabilize the drift the of bump position, which is commonly observed in conventional lateral inhibition type neural networks. We also investigated the partial annealing framework in which the interaction between neurons changes. Less