Using network models consisting of gap junction coupled Wang-Buszaki neurons, we demonstrate that it is possible to obtain not only synchronous activity between neurons but also a variety of constant phase shifts between 0 and π. We call these phase shifts intermediate stable phaselocked states. These phase shifts can produce a large variety of wave-like activity patterns in one-dimensional chains and two-dimensional arrays of neurons, which can be studied by reducing the system of equations to a phase model. The 2πperiodic coupling functions of these models are characterized by prominent higher order terms in their Fourier expansion, which can be varied by changing model parameters. We study how the relative contribution of the odd and even terms affect what solutions are possible, the basin of attraction of those solutions and their stability. These models may be applicable to the spinal central pattern generators of the dogfish and also to the developing neocortex of the neonatal rat.
Oscillations and irregular persistent firing patterns in a homogeneous network of excitatory stochastic neurons with gap junctions in the mean-field limit
Large Associative Memory Problem in Neurobiology and Machine Learning
MIT-IBM Watson AI Lab(MIT-IBM沃森人工智能实验室);IBM Research(IBM研究院);Princeton Neuroscience Institute(普林斯顿神经科学研究所);Princeton University(普林斯顿大学)
密集关联记忆或现代Hopfield网络允许存储和可靠检索指数级大(在特征空间的维度上)数量的记忆。与此同时,它们的朴素实现是非生物的,因为它似乎需要神经元之间存在多体突触连接。我们表明,这些模型是一个更微观的(书面的生物自由度)理论,有额外的(隐藏的)神经元,只需要两个身体之间的相互作用的有效描述。由于这个原因,我们提出的微观理论是一个有效的模型,大联想记忆与一定程度的生物相容性。我们的网络的动力学和它的降维等效都最小化能量(李雅普诺夫)函数。当某些动态变量(隐藏神经元)从我们的微观理论中整合出来时,可以恢复许多以前在文献中讨论过的模型,例如“Hopfield Networks is All You Need”论文中提出的模型。我们还提供了另一种推导的能量函数和更新规则中提出的上述文件,并澄清这一类的各种模型之间的关系。