STDP in lateral connections creates category-based perceptual cycles for invariance learning with multiple stimuli

Area of Science:

• Neuroscience
• Computational Neuroscience
• Computer Vision

Background:

• Object recognition in primates is challenging due to variations in viewpoint, scale, and position.
• Existing models often struggle with cluttered scenes containing multiple objects.
• The ventral visual system handles object and face recognition.

Purpose of the Study:

• To investigate a mechanism for simultaneous object segmentation and recognition in cluttered scenes.
• To explore the role of plastic lateral connections in organizing visual representations.
• To explain how the visual system overcomes challenges in naturalistic environments.

Main Methods:

• Simulations using artificial stimuli with category information encoded in plastic lateral connections.
• Utilizing spike-timing-dependent plasticity (STDP) for training feed-forward connections.
• Analyzing the organization of input representations into perceptual cycles.

Main Results:

• Experience-guided lateral interactions robustly organize input representations into perceptual cycles.
• This organization allows for the formation of independent, translation-invariant output representations.
• Demonstrated a functional role for plasticity in visual cortex lateral connectivity.

Conclusions:

• Plasticity in lateral connections is crucial for segmenting and recognizing objects in complex visual scenes.
• The proposed mechanism offers a functional explanation for how the visual system achieves robust object recognition.
• This work provides insights into the computational principles underlying visual perception.