A Double-Layer Multi-Resolution Classification Model for Decoding Spatiotemporal Patterns of Spikes With Small Sample...

Area of Science:

• Computational Neuroscience
• Machine Learning
• Systems Neuroscience

Background:

• Decoding neural activity from spike patterns is crucial for understanding brain function.
• Existing methods often struggle with high-dimensional neural data and limited sample sizes.
• Accurate decoding requires capturing spatiotemporal dynamics at various resolutions.

Purpose of the Study:

• To develop a robust classification model for decoding single-trial spatiotemporal spike patterns.
• To address challenges of high dimensionality and small sample sizes in neural decoding.
• To integrate multi-resolution temporal features for improved classification accuracy.

Main Methods:

• A double-layer ensemble classifier integrating B-spline functional expansion and L1-regularized logistic regression.
• Dimensionality reduction using B-spline expansion with varying knot numbers for multi-resolution feature extraction.
• Bootstrap aggregating (bagging) to reduce variance of base learners.
• A meta-learner in the second layer fuses multi-resolution features for final prediction.

Main Results:

• The model effectively decodes spatiotemporal spike patterns from synthetic and experimental data (rodents, humans).
• Demonstrated ability to avoid overfitting and achieve accurate predictions with small sample sizes.
• The double-layer, multi-resolution approach consistently outperformed single-layer, single-resolution classifiers.

Conclusions:

• The proposed model offers an effective solution for neural decoding from spike trains.
• Integrating multi-resolution temporal features significantly enhances decoding performance.
• This method provides a powerful tool for analyzing complex neural dynamics in cognitive tasks.