.png){kind=icon}
.png){kind=icon}
Return
Neuromorphic neural decoding model towards high-performance neuroprosthetics
.png){kind=icon}
PRE
AI
Y
T
Y
G
DOI：10.1016/j.neucom.2026.134742.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Neuroprosthetics hold exciting potential for neural rehabilitation by restoring, enhancing, or replacing the functions of the brain. However, a deep and close connection between the brain and machine-based neuroprosthetics is difficult to achieve, given the heterogeneity of representation and computing on both sides. This study demonstrates that neuromorphic model-based neuroprosthetics, mimicking the computing mechanisms of neural systems, can facilitate close and efficient interaction with the brain and, most importantly, enable rapid calibration with minimal training data, which has been a critical challenge in traditional BCIs. Specifically, we propose the synaptic plasticity-optimized spiking neural network (SpoSNN), a liquid state machine where the transmission between biological and neuromorphic neurons is optimized with synaptic learning rules shared by both sides. Results show that SpoSNN achieves competitive decoding accuracy together with low training data requirements compared to traditional neural decoders. Furthermore, we quantize the model and deploy it on a neuromorphic chip, which achieves effective on-chip neural decoding under hardware constraints. These findings suggest the potential of neuromorphic models in practical neural decoding and for brain-implantable BCI devices.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
Cited Papers