PsiZ

PsiZ infers continuous, multivariate psychological embeddings of stimuli from human similarity judgments for use in behavioral research and cognitive modeling.

Key Features:

• Flexible Trial Configurations: Supports multiple trial configurations for collecting similarity judgments, including triplet comparisons and sorting-based group assignments.
• Likelihood Model with Diverse Kernels: Implements a likelihood model that supports three classes of similarity kernels and enables parameter inference via gradient descent.
• Active Selection Algorithm: Incorporates an active selection algorithm that proposes comparisons to impose strong constraints on the embedding and reduce the number of trials required.
• Group-Specific Attention Weights: Allows specification of group-specific attention weight parameters within the likelihood model to capture perceptual differences across participant groups.

Scientific Applications:

• Behavioral research: Generates embeddings that capture human-perceived similarities among stimuli for experimental analysis.
• Cognitive psychology: Provides representational embeddings to test theories of perception and similarity-based cognition.
• Neuroscience: Produces stimulus similarity representations applicable to studies of perceptual organization and similarity in neuroscience.
• Psychometrics: Supports modeling individual- and group-level perceptual differences using attention-weight parameters.

Methodology:

Uses a likelihood model with three classes of similarity kernels, joint kernel and metric learning techniques, parameter inference via gradient descent, an active selection algorithm for proposing comparisons, and group-specific attention-weight parameters.