Jing Yang Zhou (Flatiron Institute) Gastvortrag
The perception of sensory attributes is often quantified through measurements of discriminability (an observers' ability to detect small changes in stimulus), as well as direct judgements of appearance or intensity. Despite their ubiquity, the relationship between these two measurements is controversial and unresolved. Here, we propose a framework in which they both arise from the properties of a common internal representation. Specifically, we assume that direct measurements of stimulus intensity (e.g., through rating scales) reflect the mean value of an internal representation, whereas measurements of discriminability reflect the ratio of the derivative of mean value to the internal noise amplitude, as captured by the measure of Fisher Information. Combination of the two measurements allows unique identification of internal representation properties. As a central example, we show that Weber's Law of perceptual discriminability can co-exist with Stevens' observations of power-law scaling of perceptual intensity ratings (for all exponents), when the noise amplitude increases proportionally to the internal representational mean. We extend this result by incorporating a more general physiology-inspired model for noise and a discrimination form that extends beyond Weber's range, and show that the combination allows accurate prediction of intensity ratings across a variety of sensory modalities and attributes. Our framework unifies two major perceptual measurements, provides a potential neural interpretation for the underlying representations, and discovers a new interpretation of super-threshold comparative judgement (e.g. Fechner's interpretation of Weber's law).
Jingyang Zhou did maths (mostly geometry) as an undergraduate student, and neuroimaging in grad school, both at New York University.
Now she works towards synthesizing different types of perceptual measurements under a single theoretical framework.
Her main interests include using geometrical tools to theorize human behaviors and statistical tools to interact with different types of neuronal data