Sparse coding with an overcomplete basis set: a strategy employed by V1?

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Cortex

Background:

• Simple cells in mammalian striate cortex have localized, oriented, and bandpass receptive fields.
• These properties are comparable to wavelet transform basis functions.
• Previous work suggested sparse coding explains these receptive field properties.

Purpose of the Study:

• To provide an expansive description of sparse coding for receptive field properties.
• To examine the neurobiological implications of sparse coding, particularly overcomplete codes.
• To explain non-linearities in simple cell responses using sparse coding.

Main Methods:

• Modeling receptive fields using sparse coding strategies.
• Analyzing overcomplete codes where basis functions are non-orthogonal and not linearly independent.
• Investigating deviations from linearity in input-output functions.

Main Results:

• Sparse coding successfully accounts for the localized, oriented, and bandpass receptive fields of simple cells.
• Overcomplete sparse coding leads to deviations from linearity in neural responses.
• These deviations offer a potential explanation for observed non-linearities in cortical simple cells.

Conclusions:

• Sparse coding provides a unifying framework for understanding simple cell receptive field properties.
• The model predicts specific interactions among neurons in response to naturalistic stimuli.
• This approach offers insights into the neurobiological basis of visual processing and neural computation.