Spiking Neural Network Liquid State Machine for BMI Trajectory Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current brain machine interface (BMI) systems lack an effective algorithm for decoding animal movement trajectories based on motor nerve signals using spiking neural networks, which limits their accuracy and efficiency in predicting motion information.
Innovation Solution
A brain machine interface decoding method utilizing a liquid state machine model based on spiking neural networks, where Spike-timing-dependent plasticity (STDP) is used to train connection weights without supervision, and ridge regression with supervision is applied to train readout weights, enabling real-time prediction of arm motion trajectories with improved efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep models like RNN and LSTM are used to decode neural signals, then motion signal decoding accuracy is improved, but training time and computational complexity increase
Solution Approach 1:
The patent segments the neural network into a spiking neural network component for temporal processing and a liquid state machine component for motion decoding. This segmentation allows each component to specialize in specific tasks, achieving high decoding accuracy while reducing overall training time compared to using deep RNN/LSTM models alone.
Solution Approach 2:
The patent introduces a liquid state machine as an intermediary between the spiking neural network and the final motion output. This intermediary layer processes the temporal spikes into meaningful motion representations, enabling accurate decoding without requiring the entire deep network to be trained end-to-end, thus reducing training time.
2Reliability
If spiking neural networks are used to process temporal information, then biological interpretability and anti-noise capability are improved, but lack of unified training algorithm limits practical application
Solution Approach 1:
The patent creates a universal training framework that combines STDP for unsupervised temporal feature learning with ridge regression for supervised motion decoding. This multi-functional approach allows the same network architecture to benefit from both unsupervised pre-training (improving anti-noise capability) and supervised fine-tuning (ensuring accurate motion prediction), effectively addressing the training algorithm deficiency.
Solution Approach 2:
The patent applies preliminary action by using STDP to pre-train the spiking neural network on temporal spike patterns before applying ridge regression for final motion decoding. This preliminary unsupervised training establishes robust temporal representations that are resistant to noise, and the subsequent supervised training refines these representations for accurate motion prediction.
3Productivity
If liquid state machine model with STDP is used to train connection weights, then training efficiency is improved, but decoding accuracy must be verified against deep models
Solution Approach 1:
The patent replaces the mechanical processing of deep RNN/LSTM models with a biologically-inspired spiking neural network mechanism. The STDP learning rule substitutes complex backpropagation computations with biologically plausible spike-timing-dependent weight updates, achieving faster training efficiency while maintaining decoding accuracy through the liquid state machine's natural temporal processing capabilities.
Data Source
AI summary
The present invention discloses a brain machine interface decoding method based on spiking neural network, comprising: (1) constructing a liquid state machine model based on a spiking neural network, the liquid state machine model consists of an input layer, an middle layer and an output layer, wherein, a connection weight from the input layer to the middle layer is Whh, a loop connection weight inside the middle layer is Whh, a readout weight from the middle layer to the output layer is Wyh; (2) Inputting a neuron spike train signal, and training each weight with the following strategy: (2-1) Using STDP without supervision to train the connection weight Whh from the input layer to the middle layer; (2-2) Setting the loop connection weight Whh inside the middle layer by means of distance model and random connection, and obtaining a middle layer liquid information R(t); (2-3) Using ridge regression with supervision to train the readout weight Wyh from the middle layer to the output layer, and establishing a mapping between the middle layer liquid information R(t) and the output motion information, and finally outputting a predicted motion trajectory. The present invention can quickly train a model in a relatively short time, predict the arm motion trajectory in real time, and achieve an improvement in efficiency and accuracy.