Maximal Dependence Capturing as a Principle of Sensory Processing

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain processes noisy and incomplete sensory information to recognize objects consistently.
• Current hierarchical models struggle to achieve invariant object representations despite accurately capturing input structures.
• Existing frameworks face theoretical and experimental inconsistencies in explaining robust object recognition.

Purpose of the Study:

• To propose a new principle for neural object encoding that overcomes limitations of current frameworks.
• To introduce the Maximal Dependence Capturing (MDC) principle for creating invariant object representations.
• To provide a unifying principle for sensory processing and neural computation.

Main Methods:

• Developed a computational framework based on dimension expansion and sparse coding.
• Implemented the Maximal Dependence Capturing (MDC) principle where neurons capture information-rich structural components.
• Assessed the framework's ability to achieve consistent object representations under various noisy conditions.

Main Results:

• The proposed framework achieves consistent object identity representations even with occlusion, corruption, or high noise.
• The Maximal Dependence Capturing (MDC) principle enables robust object recognition without learning corrupted forms or using deep networks.
• The framework successfully explains various observed receptive field properties of neurons.

Conclusions:

• Maximal Dependence Capturing (MDC) offers a unifying principle for sensory processing and neural object representation.
• This principle provides a more effective mechanism for achieving invariant object recognition compared to traditional hierarchical models.
• The computational framework demonstrates the feasibility of MDC in explaining neural robustness to sensory variability.