Images with harder-to-reconstruct visual representations leave stronger memory traces.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Psychology

Background:

• Memory formation is often a byproduct of perception, not just intentional selection.
• Understanding the interplay between perception and memory is crucial for cognitive architecture.
• The level-of-processing theory suggests deeper perceptual processing enhances memory durability.

Purpose of the Study:

• To investigate how perception influences memory encoding and retrieval.
• To propose and test a computational model linking image perception to memory performance.
• To determine if reconstruction error in image processing predicts memorability.

Main Methods:

• Developed a sparse coding model to compress image feature embeddings.
• Utilized an open memorability dataset of scene images.
• Performed model-driven psychophysics to validate findings.

Main Results:

• Reconstruction residuals from the sparse coding model accurately predict image memorability.
• Reconstruction error explains variance in both memory accuracy and retrieval response latencies.
• The model's predictive power for retrieval latencies surpasses that of vision-only models.

Conclusions:

• Reconstruction error serves as a key signal at the interface of perception and memory.
• This signal may influence memory through adaptive modulation of perceptual processing.
• The findings provide a computational framework for understanding perception-memory links.