Density-based clustering: A 'landscape view' of multi-channel neural data for inference and dynamic complexity...

Area of Science:

• Computational Neuroscience
• Statistical Modeling
• Network Analysis

Background:

• Analyzing multiple neural recordings presents challenges in time series representation and inferring neural network connectivity.
• Existing methods may struggle with high-dimensional neural data and complex dynamics.

Purpose of the Study:

• To demonstrate state-space clustering as an effective method for dimensionality reduction of neural time series.
• To show state-space clustering improves the inference of synaptic couplings and enables compact representation of neural dynamics.
• To apply and validate state-space clustering on both Hopfield and multi-modular spiking neural networks.

Main Methods:

• Utilized a variant of the 'mean-shift' algorithm for state-space clustering.
• Validated the approach on a Hopfield network in the glassy phase.
• Applied a Boltzmann learning-inspired procedure to a multi-modular spiking network for learning synaptic couplings.

Main Results:

• State-space clustering significantly improved the inference of synaptic couplings by providing a parsimonious parametrization of the synaptic matrix.
• Clustering effectively extracted relevant features from the state-space landscape of a spiking network.
• The cluster structure facilitated the conversion of multi-dimensional neural dynamics into a symbolic dynamics of cluster transitions.

Conclusions:

• State-space clustering offers an efficient and informative approach for analyzing neural time series and inferring network connectivity.
• This method provides a compact representation of neural dynamics, suitable for complexity measures and symbolic analysis.
• The findings highlight the utility of state-space clustering in advancing the statistical modeling of neural data.