A detection theory account of visual short-term memory for color
A detection theory account of visual short-term memory for color
复制标题
颜色视觉短期记忆的检测理论解释
作者：
P. Wilken;W. Ma
先前的研究表明，视觉短期记忆 (VSTM) 的存储限制约为 4 个项目。然而，用于推导该估计的高阈值模型（HTM）类型基于许多没有吸引力的假设，并且在其他实验范式（例如视觉搜索）中受到批评。在这里，我们报告了三个实验的结果，其中使用信号检测理论 (SDT) 方法对彩色刺激的 VSTM 进行建模。在实验 1 和 2 中，依次呈现由排列在假想圆上的彩色方块组成的两个阵列，每个阵列持续 100 毫秒，ISI 为 1500 毫秒。每个方块的颜色是从 9 种高度可辨别的颜色的调色板中随机分配的。观察者被要求以是/否的方式报告第一和第二阵列之间是否存在任何颜色差异，并以 1-4 等级对他们的反应的信心进行评分。在实验 1 中，只有一种颜色差异可能出现 (T= 1)，而组大小发生变化 (N= 2, 4, 6, 8)。在实验 2 中，集合大小是固定的 (N= 8)，而可能改变的刺激数量是变化的 (T= 1, 2, 3, 4)。针对两个实验生成的 ROC 测试了三个通用模型。除了 HTM 之外，还尝试了两种 SDT 模型：一种假设在决策之前对信号进行求和，另一种则使用最大规则。只有最大模型能够很好地拟合实验 1 和 2 的 ROC。进行了第三个实验，其中要求观察者直接报告（通过色轮）来自不同设置大小的数组（N = 2、4、6、8）的 1500 毫秒之前呈现的刺激的颜色。总的来说，我们的结果表明观察者并行独立地编码刺激，并且性能受到内部噪声的限制，内部噪声是集合大小的函数。使用 VSTM 的 HTM 报告的“4 项限制”似乎是源自渐近性能的人工产物，作为随着集合大小增加而增加的噪声的函数。
Previous studies have suggested that visual short-term memory (VSTM) has a storage limit of approximately 4 items. However, the type of high-threshold model (HTM) used to derive this estimate is based on a number of unattractive assumptions, and has been criticized in other experimental paradigms (eg, visual search). Here we report findings from three experiments in which VSTM for colored stimuli was modelled using a signal detection theory (SDT) approach. In Experiments 1 and 2, two arrays composed of colored squares arranged on an imaginary circle were presented sequentially, each for 100 ms, with a 1500 ms ISI. The color of each square was randomly assigned from a palette of 9 highly discriminable colors. Observers were asked to report in a yes/no fashion whether there were any color differences between the first and second arrays, and to rate their confidence in their response on a 1–4 scale. In Experiment 1 only one color difference could occur (T= 1) while set size was varied (N= 2, 4, 6, 8). In Experiment 2, set size was fixed (N= 8) while the number of stimuli which might change was varied (T= 1, 2, 3, 4). Three general models were tested against the ROCs generated by the two experiments. In addition to the HTM, two SDT models were tried: one assuming summation of signals prior to a decision, the other using a max rule. Only the max model was shown to provide a good fit for the ROCs of both Experiments 1 and 2. A third experiment was performed in which observers were asked to directly report (via a color wheel) the color of a stimulus presented 1500 ms previously, from an array of varying set size (N= 2, 4, 6, 8). Overall our results suggest observers encode stimuli independently in parallel, and that performance is limited by internal noise, which is a function of set size. The ‘4-item limit’reported using HTMs of VSTM appears to be an artefact derived from asymptotic performance as a function of increasing noise as set size is increased.