JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Sparse Spiking Neural-Like Membrane Systems on Graphics Processing Units

  • 0Research Group on Natural Computing, Department of Computer Science and Artificial Intelligence, I3US, SCORE Lab, Universidad de Sevilla, Avda. Reina Mercedes s/n, 41012, Sevilla, Spain.
International Journal of Neural Systems +

|

May 16, 2024

|

May 16, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study implements and parallelizes matrix compression methods for Spiking Neural P systems on GPUs. These optimized methods significantly improve simulation efficiency compared to existing GPU solutions.

Area Of Science

  • Computational Neuroscience
  • Artificial Intelligence
  • Parallel Computing

Background

  • Spiking Neural P systems are simulated using matrix representations, often leading to inefficiencies with sparse matrices.
  • Existing parallel simulation methods rely on matrix-vector multiplication, which is resource-intensive for non-fully connected neural graphs.
  • Previous compression techniques for sparse matrices in this context were proposed but lacked implementation and parallelization.

Purpose Of The Study

  • To implement and parallelize two matrix compression methods for Spiking Neural P systems on GPUs.
  • To develop a new simulator for Spiking Neural P systems with delays incorporating these compression techniques.
  • To evaluate the performance of the implemented methods against state-of-the-art GPU libraries.

Main Methods

  • Implementation of matrix compression algorithms for sparse adjacency matrices.
  • Parallelization of the compression methods and simulation on Graphics Processing Units (GPUs).
  • Development of a novel Spiking Neural P system simulator with delays and parallelized compression.

Main Results

  • The implemented and parallelized compression methods demonstrate superior performance in Spiking Neural P system simulations.
  • Significant improvements in computational resource utilization (time and memory) were observed.
  • The new simulator with compression outperformed existing solutions based on standard GPU libraries.

Conclusions

  • Matrix compression is crucial for efficient parallel simulation of sparse Spiking Neural P systems.
  • The developed GPU-based simulator with parallelized compression offers a significant advancement in performance.
  • These findings pave the way for more complex and efficient simulations of Spiking Neural P systems.

Keywords: