视知觉学习神经机制的计算模型研究

Visual perceptual learning attracts a great amount of efforts in the field. Bunch of experiments have been carried out and theories have been proposed to reveal the neural mechanism underlying the visual perceptual learning. However, the experimental results are not consistent and theories propose different neural mechanisms due to these researches focus on single cortical area or single neural process. Here, we think that visual perceptual learning is the results of plasticity of any cortical area or connectivity involving visual perceptual task. We assume that the neural circuitry involving visual perceptual learning is an intact ensemble and its parts interact with each other. The perceptual learning tasks change the parameters of the neural circuitry and the dynamic properties of the circuitry, which result in perceptual learning. But the former researches seldom investigate the circuitry as whole and ignore the dynamics of the circuitry. Thus, we will construct a large scale of visual neural network including V1, V2, V4 and LIP area to investigate the dynamics of visual circuitry by simulating the contour-detection task that has been carried out using nonhuman primates. We will explore the dynamics of the large-scale network by the variation of connectivity and the variation of single neuron’s dynamics. We will try to reveal the neural mechanism underlying visual perceptual learning by investigating the role of feedforward and feedback projections and the dynamical population coding of the visual circuitry.

视知觉学习研究领域取得了丰富的实验结果并提出多种理论模型。但实验结果相互矛盾，提出的理论模型也有分歧，原因是已有研究往往认为某个特定脑区或者特定加工环节的可塑性变化导致知觉学习。本项目认为视知觉学习中任一环节的改善都可以导致行为学的进步，参与视知觉学习的神经环路是相互作用的整体，构成高维的非线性动力系统。视知觉训练改变了动力系统的参数和动力学性质，从而提高知觉分辨能力。但是，以往的模型研究很少将视知觉环路作为整体来通盘考察，对视知觉学习的神经环路的动力学性质也不够了解。为此，本项目将以神经环路相对清楚且已有丰富电生理数据的轮廓检测的知觉学习为突破口，建立包含V1, V2,V4和LIP区域的大规模神经元网络模型，模拟猕猴轮廓检测的知觉学习过程，研究视知觉学习神经环路动力学行为的演化，考察各脑区和脑区间的连接在视知觉学习中的作用，从群体编码和神经环路的动力学性质出发，揭示视知觉学习的神经机制。

本项目聚焦视知觉学习的神经机制，通过多种数据降维的方法分析了实验记录的神经放电数据，发现视皮层神经元在轮廓线检测知觉训练过程中逐渐形成了表征有轮廓和无轮廓两个类别的活动模式，从而实现对轮廓线的高效检测。项目结合已有文献和生理基础建立视皮层的大规模神经网络模型，模拟发现初级视皮层对轮廓线刺激的响应模式受到皮层神经元横向远距离连接、高级皮层的反馈连接的重要影响。项目研究了感觉区和决策区之间相互投射对知觉决策的影响，发现相互投射通过调控单一感觉和多感觉刺激对决策脑区的有效输入影响决策阈值，使网络整合的结果既可以优于也可以劣于贝叶斯最优整合。增强兴奋性的前馈和抑制性的反馈投射使网络整合优于贝叶斯整合。研究发现振荡系统一定包含扩张性和收缩性变量，且前者的相位领先后者，从而解释了长期观察到的神经振荡中兴奋性神经元相位领先抑制性神经元的现象。项目也从平行环路之间的抑制和去抑制机制解释了多巴胺神经元和外侧缰核神经元对意外结果的短时放电现象，进一步揭示了意外收益和损失引起的学习率差异导致不同的风险态度和概率感知扭曲。项目还展示了双人一致决策和动作同步的动力学过程等相关工作。

内容获取失败，请点击重试