Anthropic Correction of Information Estimates and Its Application to Neural Coding

Area of Science:

• Neuroscience
• Information Theory
• Computational Neuroscience

Background:

• Information theory, particularly mutual information (MI), is crucial for understanding neural processing.
• Estimating MI is challenging due to limited stimuli and abundant neural response data, leading to asymmetric estimation problems.

Purpose of the Study:

• To develop and evaluate novel methods for accurate mutual information estimation in neuroscience.
• To address the negative bias in direct plug-in estimators and introduce a complementary positive bias estimator.

Main Methods:

• Analysis of asymmetric information estimation problems in neurophysiological recordings.
• Introduction of an 'anthropic' correction to counteract the negative bias of direct plug-in estimators.
• Derivation of tail and variance bounds for both estimators.

Main Results:

• Direct plug-in information estimates exhibit a negative bias.
• The proposed 'anthropic' correction introduces a positive bias.
• Both estimators and their combinations are suitable for neuroscience applications.

Conclusions:

• The developed methods offer improved accuracy for information estimation in neuroscience.
• These techniques enhance the analysis of neural discrimination and redundancy, particularly in sensory systems like the avian auditory system.