Estimation of perceptual scales using ordinal embedding.

Area of Science:

• Psychology
• Machine Learning
• Psychophysics

Background:

• Measuring subjective experience (sensation) is challenging.
• The method of triads uses relative judgments (e.g., 'Is Si more similar to Sj or Sk?').
• Existing methods like NMDS and MLDS have limitations in analyzing these judgments.

Purpose of the Study:

• To propose and evaluate ordinal embedding methods for estimating sensation scaling functions from relative judgments.
• To compare the performance of ordinal embedding against established psychophysical methods.
• To enable the estimation of multidimensional perceptual scales.

Main Methods:

• Utilizing ordinal embedding techniques from machine learning.
• Applying these methods to analyze data from the method of triads.
• Comparing performance through extensive simulations and real psychophysics experiments.
• Reviewing nonmetric multidimensional scaling (NMDS) and maximum likelihood difference scaling (MLDS).

Main Results:

• Ordinal embedding effectively estimates scaling functions from fewer relative judgments compared to existing approaches.
• This method can successfully estimate multidimensional perceptual scales.
• Performance is comparable to MLDS in one dimension and superior in higher dimensions.
• An R-implementation is provided for accessibility.

Conclusions:

• Ordinal embedding offers a powerful and flexible approach for analyzing subjective sensory experiences.
• It advances the field of psychophysics by enabling more robust and multidimensional scale estimation.
• The proposed methods are practical and accessible for researchers.