Spikebench: An open benchmark for spike train time-series classification.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Data Science

Background:

• Neural decoding relies on machine learning models like decision trees and deep neural networks.
• Analyzing neural spike trains, a form of time-series data, requires robust and diverse benchmarks.
• Existing benchmarks may not fully capture the complexity of neural decoding tasks.

Purpose of the Study:

• To propose a novel spike train classification benchmark for neural decoding.
• To evaluate the effectiveness of traditional feature engineering against deep learning models.
• To provide reproducible resources for advancing neural decoding research.

Main Methods:

• Development of a benchmark using open-access neural activity datasets.
• Inclusion of diverse learning tasks: stimulus classification, behavioral state prediction, neuron identification.
• Comparison of hand-crafted time-series feature engineering with state-of-the-art deep learning models.

Main Results:

• The proposed benchmark effectively evaluates neural decoding algorithms.
• Hand-crafted feature engineering achieved performance comparable to deep learning models.
• This demonstrates the strength of traditional methods as a baseline for neural decoding.

Conclusions:

• A new benchmark for spike train classification is established, facilitating rigorous evaluation of neural decoding methods.
• Traditional feature engineering provides a competitive baseline, highlighting its continued relevance.
• The released code enables reproducibility and further research in the field.