科学与工程计算系列讨论班—Modeling within and across area neuronal variability in the visual system
主 题: 科学与工程计算系列讨论班—Modeling within and across area neuronal variability in the visual system
报告人: 黄橙橙 博士 (University of Pittsburgh)
时 间: 2017-06-15 15:30-16:30
地 点: 理科1号楼1418
Abstract: Shared variability among neurons (noise correlations) have been commonly observed in multiple cortical areas (Cohen and Kohn, 2011). Moreover, noise correlations are modulated by cognitive factors, such as overall arousal, task engagement and attention (Cohen and Maunsell, 2009; Doiron et al. 2016). While there is much discussion about the consequences of noise correlations on neuronal coding, there is a general lack of understanding of the circuit mechanisms that generate and modulate shared variability in the brain. Recently, simultaneous microelectrode array recordings from V1 and MT in behaving monkeys (Ruff and Cohen, 2016) suggest that attention not only decreases correlations within a cortical area (MT), but also increases correlations between cortical areas (V1 and MT). The differential modulation of between-areas and within-area noise correlations impose further constraints on circuit mechanisms for the generation and propagation of noise correlations. We develop a spiking neuron network with spatiotemporal dynamics that internally generates shared variability matching the low dimensional structure widely reported across cortex. The low dimensional variability requires slow and narrow inhibition, which are consistent with physiology. We further show that the internally generated variability results from macroscopic chaos in population rates, which correlates neurons from the same recurrent network while decoupling them from feedforward inputs. Attention is modeled as depolarizing the inhibitory neuron population, which reduces the internally generated shared variability and allows the network to better track input signal. Our model provides a much needed mechanism for how shared variability is both generated and modulated in recurrent cortical networks.