Summary
In Bayesian framework to perception, all fixation of perceptual belief is assumed to be connected to the computation of Bayesian posterior probability. Bayesian inference is a statistical procedure, that results in an optimal combination of the available evidence with prior beliefs. In perception, generally, this approach entails a rational estimate of the structure of the scene that combines fit to the available image data with the mental set of the perceiver (background knowledge, context, etc.). The ultimate purpose of the project is to test quantitatively the necessary and sufficient conditions (“If A, then B” and “A, only if B”) of the Bayesian framework for the occurrence of the new phenomena under consideration and, in the light of the results, to enrich and improve Bayesian framework.Description
To reach the project goal, the general methods are based on threefold procedures: i) experimental phenomenology, aimed to explore qualitatively the main attributes of the illusions; ii) psychophysical methods useful to measure quantitatively the prior probability p(H) and knowledge specified by the likelihood, p(D|H); iii) Bayesian ideal observer, designed to maximize a performance measure for a visual task (e.g., proportion of correct decisions) and, as a result, useful as a benchmark with which to compare human performance for the same task. Ideal observers provide the strongest psychophysical test since they are complete models of visual performance based on both the posterior and the task. ‘Bayesian psychophysics’ typically proceeds from the definition of a simple experimental task and the specification of how a Bayesian observer would perform in that task. The experimental task is such that experimenters can determine probability distributions necessary to test whether subjects’ performance is consistent with Bayesian inference.
They include three main classes crucial for our purposes:
- similarity-dissimilarity among components within the same figures eliciting what has been called accentuation principle, actually representing an antinomy for the Bayesian approach;
- a new apparent motion illusion, which further enhances the antinomy due to the accentuation principle and
- invisible paradoxes of shadows vs. shading vs. illumination and pseudo-shadows illusions revealing contradictions between potential priors and likelihood.
The main phenomenal outcomes of the stimuli are described within the legends included in each figure. Since the attractive feature of Bayesian models of behavior is that they provide descriptions of what would be optimal for a given task, thus, they are often referred to as ‘ideal observer’ models because they quantify how much to update our beliefs in light of new evidence. Departures from these normative models can then be explained in terms of other constraints such as computational complexity or individual or developmental differences. This entails the need within this project to test different classes of subjects belonging to different ages (from 5 to 25 years old, going through the main critical developmental ages), gender, education level, etc.