An Infomax algorithm can perform both familiarity discrimination and feature extraction in a single network

Area of Science:

• Neuroscience
• Cognitive Psychology
• Computational Neuroscience

Background:

• The brain's capacity for discriminating visual stimuli as novel or familiar is vast.
• The perirhinal cortex is crucial for familiarity discrimination and feature extraction.
• Current models struggle to perform both feature extraction and familiarity discrimination for numerous stimuli simultaneously.

Purpose of the Study:

• To investigate if a visual feature extraction model can simultaneously perform familiarity discrimination.
• To assess the capacity of such a model for processing large numbers of stimuli.
• To explore the relationship between feature extraction and familiarity discrimination in neural models.

Main Methods:

• Utilized the Infomax model, a known visual feature extraction model.
• Evaluated the model's ability to perform both feature extraction and familiarity discrimination.
• Compared the model's capacity with previously proposed models, especially under correlated input conditions.

Main Results:

• The Infomax model efficiently performed both feature extraction and familiarity discrimination simultaneously.
• The model demonstrated a significantly larger capacity compared to prior models, particularly with correlated inputs.
• Feature extraction by the model markedly enhanced its subsequent familiarity discrimination ability.

Conclusions:

• A single model (Infomax) can effectively handle both feature extraction and familiarity discrimination.
• This model offers a potential explanation for perirhinal cortex functions.
• The findings suggest a strong link between detailed feature representation and familiarity assessment in neural systems.