Aberdeen Medical School, Institute of Medical Sciences, Aberdeen, Scotland AB25 2ZD, United Kingdom.
Chaos. 2010 Dec;20(4):045118. doi: 10.1063/1.3524305.
Human sensory processing can be viewed as a functional H mapping a stimulus vector s into a decisional variable r. We currently have no direct access to r; rather, the human makes a decision based on r in order to drive subsequent behavior. It is this (typically binary) decision that we can measure. For example, there may be two external stimuli s([0]) and s([1]), mapped onto r([0]) and r([1]) by the sensory apparatus H; the human chooses the stimulus associated with largest r. This kind of decisional transduction poses a major challenge for an accurate characterization of H. In this article, we explore a specific approach based on a behavioral variant of reverse correlation techniques, where the input s contains a target signal corrupted by a controlled noisy perturbation. The presence of the target signal poses an additional challenge because it distorts the otherwise unbiased nature of the noise source. We consider issues arising from both the decisional transducer and the target signal, their impact on system identification, and ways to handle them effectively for system characterizations that extend to second-order functional approximations with associated small-scale cascade models.
人类的感官处理可以被看作是一个功能 H，它将刺激向量 s 映射到决策变量 r 上。我们目前无法直接访问 r；相反，人类根据 r 做出决策，以驱动后续行为。我们可以测量的正是这种（通常是二进制的）决策。例如，可能有两个外部刺激 s([0]) 和 s([1])，由感官设备 H 映射到 r([0]) 和 r([1])；人类选择与最大 r 相关联的刺激。这种决策转换对 H 的准确描述构成了重大挑战。在本文中，我们探索了一种基于反向相关技术的行为变体的特定方法，其中输入 s 包含一个目标信号，该信号被受控的噪声干扰所污染。目标信号的存在带来了额外的挑战，因为它扭曲了噪声源原本无偏的性质。我们考虑了来自决策换能器和目标信号的问题，它们对系统识别的影响，以及如何有效地处理它们，以便将系统特征扩展到具有相关小尺度级联模型的二阶功能逼近。
