Tight data-robust bounds to mutual information combining shuffling and model selection techniques.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Information Theory

Background:

• Estimating information in neural spike trains is vital for understanding brain function.
• Limited experimental sampling introduces upward bias, complicating accurate information estimation.
• Existing methods struggle with bias, particularly in noise entropy calculations.

Purpose of the Study:

• To develop novel, less biased estimators for information carried by neural spike times.
• To introduce a nonparametric test for assessing information encoding within low-dimensional response models.
• To establish data-robust upper and lower bounds for mutual information in neural data.

Main Methods:

• Utilized data-shuffling techniques to cancel bias in noise entropy estimation.
• Employed a nonparametric test to evaluate information decoding under low-dimensional assumptions.
• Combined bias reduction and low-dimensional modeling for precise information quantity estimation.

Main Results:

• Developed a new information estimator with significantly reduced bias and comparable variance to standard methods.
• Demonstrated that complex neural response spaces can be captured by a few parameters.
• Achieved tight, data-robust bounds on mutual information, effective even with minimal trials per stimulus.

Conclusions:

• The novel methods provide precise estimators for neural information, overcoming limitations of traditional approaches.
• These techniques enable accurate quantification of information in spike trains, even with strong neural correlations.
• The findings are validated through simulations and in vivo recordings from the somatosensory cortex.