An Information-Geometric Formulation of Pattern Separation and Evaluation of Existing Indices

Area of Science:

• Computational neuroscience
• Information geometry
• Neural coding

Background:

• Pattern separation is crucial for distinguishing similar inputs in neural systems.
• Existing computational models often simplify the complex relationships between neural activity patterns.
• Understanding how neural ensembles encode information requires robust methods to analyze spike train similarity.

Purpose of the Study:

• To develop an information-geometric framework for modeling pattern separation.
• To implement and analyze a two-neuron system to evaluate existing spike train similarity indices.
• To identify limitations of current metrics in capturing neural pattern differences, particularly those related to spike timing and correlation.

Main Methods:

• Formulating pattern separation using statistical distributions on a manifold.
• Implementing a two-neuron system with a probability law defining a 3D manifold.
• Analyzing the sensitivity of common spike train similarity indices to marginal and correlational firing rates.

Main Results:

• The information-geometric model demonstrates pattern separation as generating dissimilar neural patterns from similar inputs.
• Most tested similarity indices are sensitive to marginal firing rate differences.
• No index adequately captures differences in spike trains arising from altered neural activity correlation or relative spike timing.

Conclusions:

• Current spike train similarity indices are insufficient for fully characterizing pattern separation.
• Metrics need refinement to account for the role of precise spike timing and neural synchrony.
• The proposed information-geometric framework offers a novel perspective for analyzing neural coding and pattern separation.