The Markovian and Memoryless Properties of Visual System: Evidence from Serial Face Processing

TL;DR

This study investigates whether the visual system exhibits Markovian or memoryless properties during serial face processing, finding that these properties are violated under certain conditions, indicating complex system dynamics.

Contribution

It provides empirical evidence that the visual system's serial dependence can violate Markovian and memoryless assumptions, revealing asymmetrical properties across different facial traits.

Findings

Serial dependence was absent in certain conditions.

Markovian and memoryless properties were violated in the two-trait condition.

Different facets of the same task may follow asymmetrical system properties.

Abstract

The visual system can be viewed and studied as an information processing system. If so, then the visual system should follow specific fundamental properties: either a memory or a memoryless system. Previous studies in serial dependence in vision found that the perception of the current stimulus is positively determined by the previous one. However, we are not entirely sure whether this phenomenon is a Markov processing. In this study, participants were asked to rate the social characteristics (attractiveness, trustworthiness, and dominance) of a face, either followed by the same characteristic (the one-trait condition) or another one (the two-trait condition) in randomized orders. By doing so, we can directly test the contribution of the previous input and output to the current output and thus study the properties of the system. Using Derivative of Gaussian, Markov Chain and Linear Mixed effect modeling, convergent results suggested that the serial dependence was absent and the memoryless and Markovian properties were violated in the two-trait condition when testing both attractiveness and dominance, but not in the other conditions. Thus, different facets of (presumably) the same computational task may follow asymmetrical system properties. The study also develops serial dependence as an effective technique to reveal the relationships between different computation tasks.