Related Experiment Video
Updated: Jun 18, 2026
13:40
Examining Local Network Processing using Multi-contact Laminar Electrode Recording
Published on: September 8, 2011
12.8K
Surrogate gradient learning in spiking networks trained on event-based cytometry dataset.
Optics Express
|June 11, 2024
Summary
This study integrates spiking neural networks (SNNs) with event-based vision sensors for advanced micro-particle classification in flow cytometry. The new method achieves over 99% accuracy for four cell types, improving upon previous models.
Area of Science:
- Biomedical Engineering
- Computational Neuroscience
- Machine Learning
Background:
- Spiking neural networks (SNNs) mimic biological brain function.
- Event-based vision sensors replicate biological eye mechanisms.
- Previous work achieved >98% accuracy using logistic regression for binary cell classification.
Purpose of the Study:
- Integrate SNNs and event-based vision sensors for micro-particle classification.
- Improve classification accuracy for a wider variety of cells with subtle morphological differences.
- Address challenges in training SNNs, such as vanishing gradients.
Main Methods:
- Utilized spiking neural networks (SNNs) integrated with event-based vision sensors.
- Applied the surrogate gradient method to overcome vanishing gradient issues during SNN training.
- Analyzed internal network dynamics to enhance performance.
Main Results:
- Achieved over 99% classification accuracy on test data for a four-class micro-particle problem.
- Successfully trained SNNs using the surrogate gradient method.
- Enhanced network accuracy and sparsity by exploring internal dynamics.
Conclusions:
- The integration of SNNs and event-based vision sensors offers a powerful approach for label-free flow cytometry.
- The surrogate gradient method effectively enables SNN training for complex classification tasks.
- Analyzing SNN internal dynamics can further optimize classification performance and efficiency.
