Temporal spiking generative adversarial networks for heading direction decoding

Jiangrong Shen ¹, Kejun Wang ², Wei Gao ³

⁰School of Computer Science and Technology, Xi'an Jiaotong University, China; State Key Lab of Brain-Machine Intelligence, Zhejiang University, China; College of Computer Science and Technology, Zhejiang University, China.

Neural Networks : the Official Journal of the International Neural Network Society +

|

December 18, 2024

View abstract on PubMed

Summary

This study introduces Temporal Spiking Generative Adversarial Networks (T-SGAN) to create synthetic neural data for the ventral intraparietal area (VIP). This approach enhances heading direction decoding accuracy using energy-efficient spiking neural networks (SNNs).

Area Of Science

Neuroscience
Computational Neuroscience
Artificial Intelligence

Background

Spike-based neuronal responses in the ventral intraparietal area (VIP) show complex dynamics, challenging neural decoding due to limited biological data.
Collecting sufficient VIP neuronal response data for sophisticated models is practically difficult.

Purpose Of The Study

To develop a unified, energy-efficient spiking neural network (SNN) framework for generating synthetic VIP neuronal data and decoding heading direction.
To address data limitations in VIP neural decoding using generative models.

Main Methods

Proposed Temporal Spiking Generative Adversarial Networks (T-SGAN), a spiking transformer-based model, to generate synthetic time-series neuronal data.
Incorporated temporal segmentation and spatial self-attention in T-SGAN for efficient data generation.
Employed a recurrent SNN decoder with an attention mechanism for heading direction decoding.

Main Results

T-SGAN successfully generated realistic synthetic VIP neuronal response data.
The SNN-based decoding framework achieved up to a 1.75% improvement in decoding accuracy.
Demonstrated the energy efficiency of the SNN framework for neural decoding applications.

Conclusions

The proposed T-SGAN framework effectively overcomes data limitations in VIP neural decoding.
Spiking neural networks offer a promising, energy-efficient solution for complex neural decoding tasks.
The framework significantly enhances heading direction decoding accuracy using generated synthetic data.

Keywords:

Heading direction decoding Spiking generative adversarial networks Spiking neural networks