PPM-Decay: A computational model of auditory prediction with memory decay.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Auditory Perception

Background:

• Statistical learning and probabilistic prediction are key to auditory cognition.
• The Prediction by Partial Matching (PPM) model is a prominent computational tool but has limitations.
• PPM's perfect memory contradicts human cognitive constraints like memory decay.

Purpose of the Study:

• To address PPM's limitations by introducing a memory decay mechanism.
• To develop a more cognitively plausible model for auditory statistical learning.
• To enhance predictive performance in dynamic auditory sequences.

Main Methods:

• Developed PPM-Decay, a variant of PPM with a customizable memory decay kernel.
• Evaluated PPM-Decay using artificial sequences, Western music chords, and new behavioral data.
• Assessed the model's ability to capture changing statistics and memory effects.

Main Results:

• PPM-Decay demonstrated improved predictive performance on sequences with changing statistics.
• The model successfully captured the impact of memory constraints on auditory pattern detection.
• The decay kernel enhanced the model's ability to reflect human cognitive processes.

Conclusions:

• PPM-Decay offers a more accurate computational model of auditory statistical learning.
• The model's memory decay feature better represents human cognitive limitations.
• An open-source R package (ppm) is available for the PPM-Decay model.