Learning delays through gradients and structure: emergence of spatiotemporal patterns in spiking neural networks
2024

Learning Delays in Spiking Neural Networks

publication Evidence: high

Author Information

Author(s): Mészáros Balázs, Knight James C., Nowotny Thomas

Primary Institution: University of Sussex, Brighton, United Kingdom

Hypothesis

Can combining delay learning with dynamic pruning improve the efficiency of spiking neural networks for temporal data processing?

Conclusion

The study demonstrates that combining delay learning with dynamic pruning enhances the performance and efficiency of spiking neural networks.

Supporting Evidence

  • The dynamic pruning approach outperformed per-synapse delay learning in sparse networks.
  • Training introduced structural features within the network's receptive fields.
  • Significant differences in spatial correlations were observed between trained and untrained networks.

Takeaway

This study shows that teaching computers to learn from time delays in their connections can help them work better, especially when they also learn to cut unnecessary connections.

Methodology

The study used a spiking neural network model with learnable synaptic delays and dynamic pruning, evaluated on the Raw Heidelberg Digits benchmark.

Potential Biases

Potential biases may arise from the specific benchmarks and methods used for evaluation.

Limitations

The study primarily focuses on specific datasets and may not generalize to all types of spiking neural networks.

Statistical Information

P-Value

3.88 × 10−21

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.3389/fncom.2024.1460309

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication