A dynamic spatiotemporal normalization model captures perceptual and neural effects of spatial and temporal context

Area of Science:

• Neuroscience
• Computational Vision
• Perception

Background:

• The visual system processes dynamic inputs influenced by spatial and temporal context.
• Divisive normalization models spatial or temporal context separately.
• Existing models do not explain how future stimuli suppress past ones.

Purpose of the Study:

• Introduce a dynamic spatiotemporal normalization model (DSTN).
• Investigate DSTN's ability to capture bidirectional temporal context effects on neural responses and behavior.
• Unify spatial and temporal normalization within a single framework.

Main Methods:

• Developed a dynamic spatiotemporal normalization model (DSTN) with a unified spatiotemporal receptive field.
• Implemented temporal normalization via excitatory and suppressive drives with distinct temporal windows.
• Evaluated DSTN against empirical observations of neural and behavioral data.

Main Results:

• DSTN generated biphasic temporal receptive fields, aligning with empirical findings.
• The model successfully reproduced neural response properties like surround suppression, nonlinear dynamics, subadditivity, adaptation, and backwards masking.
• DSTN captured stimulus contrast-dependent bidirectional temporal suppression, matching human behavioral data.

Conclusions:

• A unified spatiotemporal normalization computation can underlie dynamic stimulus processing and perception.
• DSTN provides a cohesive framework for understanding visual context effects.
• The model demonstrates the importance of integrating spatial and temporal dimensions in visual processing.