Information-theoretic signal detection theory.

Area of Science:

• Cognitive Psychology
• Mathematical Psychology
• Information Theory

Background:

• Signal detection theory (SDT) is a standard framework for understanding stimulus classification.
• Current SDT methods often rely on specific distributional assumptions (e.g., Gaussian) and can be computationally intensive.

Purpose of the Study:

• To introduce information-theoretic tools for approximating key quantities in SDT.
• To develop a more flexible and computationally tractable framework for analyzing classification performance.

Main Methods:

• Utilized information theory to derive a lower bound on correct classification probability.
• Quantified this bound using information-theoretic properties: prior uncertainty, class separability, and model discrepancy.

Main Results:

• The proposed bound offers a simple approximation to classification performance.
• The framework generalizes to multi-dimensional stimuli and multiple categories.
• It does not require Gaussian distributions, offering broader applicability.

Conclusions:

• The information-theoretic signal detection theory (IT-SDT) framework provides new insights into classification performance.
• IT-SDT highlights the impact of discrepancies between an observer's model and environmental realities on performance.