Related Experiment Video
Updated: Jan 9, 2026
09:42
Stimulus-specific Cortical Visual Evoked Potential Morphological Patterns
Published on: May 12, 2019
6.4K
Temporally-Varying Stimulations for Cortical Visual Neuroprosthetics using Spiking Neural Networks
Summary
This study explores using spiking neural networks (SNNs) for visual neuroprostheses, enabling dynamic stimulation patterns. SNNs offer comparable results to artificial neural networks (ANNs) with reduced computational cost, paving the way for advanced visual restoration devices.
Area of Science:
- Neuroscience
- Computer Science
- Biomedical Engineering
Background:
- Visual neuroprostheses aim to restore sight through electrical stimulation.
- Current systems often use artificial neural networks (ANNs) with fixed stimulation parameters.
- There is a need for more dynamic and efficient stimulation control methods.
Purpose of the Study:
- To investigate the feasibility of using spiking neural networks (SNNs) for visual neuroprosthesis.
- To develop time-varying stimulation patterns for improved visual perception.
- To compare the performance and computational efficiency of SNNs against ANNs.
Main Methods:
- Developed an SNN-based encoder for visual neuroprosthesis.
- Utilized a phosphene simulator to model perceptual responses to SNN stimulation patterns.
- Trained and evaluated the SNN on the MNIST and N-MNIST datasets.
Main Results:
- The SNN-based encoder achieved reasonable image reconstruction quality on MNIST, comparable to state-of-the-art ANNs.
- The SNN required 2x fewer operations (1.16 GFLOPs) compared to the ANN (2.44 GFLOPs).
- Preliminary results on N-MNIST demonstrated recognizable digit shapes, suggesting potential for event cameras.
Conclusions:
- Spiking neural networks are a feasible alternative to ANNs for visual neuroprosthesis.
- SNNs offer significant computational advantages and enable dynamic stimulation patterns.
- Event cameras show promise for future visual prosthesis applications.
