Active mutual conjoint estimation of multiple contrast sensitivity functions.

Area of Science:

• Vision science
• Machine learning
• Statistical modeling

Background:

• Classical parametric methods for contrast sensitivity function (CSF) estimation have limitations in accuracy and efficiency.
• Recent nonparametric approaches offer improved tradeoffs but can be further optimized.

Purpose of the Study:

• To introduce a novel kernel and generalized trial selection for machine learning-based CSF estimation using Gaussian processes.
• To enable simultaneous estimation of multiple CSFs through latent variable representations.

Main Methods:

• Developed a flexible kernel for CSF estimation, surpassing traditional functional forms.
• Implemented generalized trial selection beyond pure information gain for data acquisition.
• Introduced latent variable representations for estimating multiple CSFs concurrently.

Main Results:

• The novel kernel, despite its flexibility, leads to a more efficient CSF estimator.
• Generalized trial selection enhances estimator quality compared to traditional methods.
• Simultaneous estimation of multiple CSFs (e.g., across eyes, eccentricities, luminances) is demonstrated.

Conclusions:

• The proposed machine learning framework offers significant improvements in CSF estimation accuracy and efficiency.
• The developed methods provide greater flexibility and enable simultaneous multi-CSF estimation.
• These advancements advance the field of vision science and psychophysical measurement.