Enhanced EEG Data Classification with 3D SNN and On-Line Learning ESN
summary
Hierarchical model maps 14-channel EEG (60 one-second trials, 128 time bins) through a NeuCube 3D spiking layer with STDP synapses whose per-neuron firing rates feed an Echo State Network reservoir of sparse tanh units that learns on-line to classify three movement classes. Grid search explores 243,000 hyperparameter combinations on an HPC, largest runs taking ~8 days (15x speedup vs non-HPC), and test error drops to about 1×10^-5, yielding near-perfect class traces when decoded for an exoskeleton control target.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:10Project goals
- 1:52Model architecture
- 3:20Biological similarity
- 5:12Firing rates and classification
- 6:19Parameter optimization
- 7:22Performance results
- 8:50Simulation time reduction with HPC
- 9:13Acknowledgements
- 9:34Discussion and audience questions
- 10:18Research goals and data
- 11:42Applications and limits
- 12:27Closing
related talks
09mNeuroAssist: Enhancing Cognition Computer Synergy with Adaptive AI & Enhance Neural Decoding
11mConnected brains
14mEEG Topologies with Neo4j: Interpolation-Driven Graph Learning for Chronic Pain Detection
05mA Feature Adaptive Learning Method for High-Density sEMG-Based Gesture Recognition
47mFunctional Triplet Motifs Underlie Accurate Predictions of Single-Trial Responses...
21mFramework for Electroencephalography-based Evaluation of User Experience
35mTranslating Brain Electrical Signals into Actions through Models...
36mU-Net-style neural networks for feature identification in 1D time-series