Modulatory feedback determines attentional object segmentation in a model of the ventral stream

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Visual cortex processing

Background:

• Neuroscience and artificial neural networks (ANNs) have a reciprocal relationship, with ANNs explaining neural tuning in the visual cortex.
• The role of modulatory feedback connections in attention and perceptual organization remains largely unresolved.

Purpose of the Study:

• To present a biologically plausible neural network model for scene segmentation and attention shifting.
• To investigate the role of modulatory feedback connections in visual processing.

Main Methods:

• Developed a neural network model incorporating feedback connections from higher to lower cortical areas.
• Simulated scene segmentation and attention shifting mechanisms.

Main Results:

• The model replicated neurophysiological signatures of recurrent processing, showing enhanced activity in figural regions compared to background.
• Modulation of activity by figure/ground occurred with a delay, dependent on feedback loops.
• Object-based attention amplified figural response enhancement without spilling over to the background, mirroring visual cortex observations.

Conclusions:

• The study demonstrates how ANNs can provide insights into recurrent cortical processing for scene segmentation.
• The model highlights the functional role of feedback connections in object-based attention and perceptual organization.