Describing complex cells in primary visual cortex: a comparison of context and multifilter LN models

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Vision Science

Background:

• Receptive field (RF) models are crucial for understanding neural responses to sensory input.
• Current popular RF models include multifilter linear-nonlinear (LN) models and context models, each with inherent assumptions and limitations.
• Different models capture distinct stimulus-response mappings, varying in their accuracy for real neural data.

Purpose of the Study:

• To introduce novel methods for estimating context models incorporating realistic logistic and exponential nonlinearities.
• To compare the performance of context models against multifilter LN models in describing neural data from complex cells.
• To evaluate the effectiveness of different receptive field modeling frameworks for complex cells in the primary visual cortex.

Main Methods:

• Estimation of context models with logistic and exponential nonlinearities.
• Evaluation of context models and multifilter LN models using recorded data from complex cells in cat primary visual cortex.
• Performance assessment based on single-spike information and correlation coefficients.

Main Results:

• Context models demonstrated superior performance compared to multifilter LN models of equivalent complexity (parameter count).
• Novel context models achieved the most significant performance improvements.
• Results indicate that the multifilter LN-model framework is suboptimal for complex cells, while the context-model framework is superior.

Conclusions:

• Context models, particularly the novel variants, provide a more accurate and interpretable description of complex cell behavior than multifilter LN models.
• The context-model framework is recommended as a superior alternative for modeling complex cells.
• The study highlights the limitations of traditional LN models and the advantages of advanced context-based approaches in neuroscience.