NEURAL CODING OF VISUAL STIMULI IN THE AWAKE MONKEY

为了研究视觉感知的机制，我们记录了 神经节细胞纤维的活动，这些纤维将视觉信号 从视网膜到大脑的信息，并将其与 记录外侧膝状体核中的单个神经元， 初级视觉皮层，前两个视网膜外阶段的视觉 下颞叶皮层，最后的视觉处理， 大脑皮层的处理站 三组神经元 记录都显示出不同的时间反应模式， 不同的视觉刺激模式。 当分析神经元时， 作为传达视觉信息的交流渠道， 刺激在他们的反应，所看到的反应模式只能 可以表示为几个（3-6）同时的总和， 独立的活动模式。 其中三项活动 模式被分析为一个时间代码，这个代码被发现， 包含的信息量是 反应强度，神经元反应的通常测量。 然而，神经节细胞纤维，携带尽可能多的 信息作为神经元在随后的阶段，基本上 时间调制码中的信息比 响应强度代码。 因此，时间调制携带 视觉信息的更大比例， 视网膜 传统上，人们认为信息 关于多个刺激参数，例如亮度，图案， 和持续时间，必须混淆在神经元 应答 然而，基于这种多路复用滤波器假设， 初级视皮层神经元反应的新分析 发现了关于每一个的信息 在响应中单独携带参数。 的几何 对数据的分析显示了神经元的潜在结构， 代码. 当使用三维空间来表示 反应，对单一模式的反应似乎在于 无论亮度或持续时间如何， 因为不同的图案经常是可分离的。 方程 描述了一种神经编码

To study the mechanisms underlying visual perception we recorded the activity of ganglion cell fibers, the fibers that send visual information from the retina to the brain, and compared it with the recordings of single neurons in the lateral geniculate nucleus and primary visual cortex, the first two extraretinal stages of visual processing, and inferior temporal cortex, the last visual processing station in the cortex. The three sets of neuronal recordings all showed different temporal response patterns to different visual stimulus patterns. When neurons were analyzed as communication channels carrying information about visual stimuli in their responses, the response patterns seen could only be represented as the sum of several (3-6) simultaneous, independent patterns of activity. Three of these activity patterns were analyzed as a temporal code, and this code was found to contain twice as much information as that conveyed by the response strength, the usual measure of neuronal response. However, the ganglion cells fibers, which carry as much information as the neurons in subsequent stages, have substantially less information in the temporal modulation code than in the response strength code. Thus, temporal modulation carries a significantly greater proportion of visual information outside of the retina. Traditionally, it has been thought that information about multiple stimulus parameters, such as luminance, pattern, and duration of presentation, must be confounded in the neuronal responses. However, based on this multiplex-filter hypothesis, a new analysis of the neuronal responses of primary visual cortex led to the discovery that information about each of these parameters is carried separately in the response. A geometrical analysis of the data shows a potential structure for a neural code. When a 3-dimensional space is used to represent the responses, the responses to a single pattern appear to lie in a single plane regardless of luminance or duration, with the planes for different patterns frequently being separable. The equations for these planes describe a neural code.