AsmaSbeih/Architectural-Transfer-from-Non-Invasive-EEG-to-Invasive-sEEG-for-Neural-Speech-Decoding: Neural Speech Decoding (EEG to sEEG Transfer)

This release contains the code and experiments for the paper: "Architectural Transfer from Non-Invasive EEG to Invasive sEEG for Neural Speech Decoding" Overview This project investigates the transfer of deep learning architectures from non-invasive EEG to invasive sEEG for neural speech decoding. Key Features Hybrid architecture integrating Multi-Receptive Residual Blocks (MR-ResBlocks) Progressive fine-tuning strategy Multi-subject support with heterogeneous electrodes Results PCC improvement: 0.288 → 0.400 MSE reduction: 4.567 → 4.194 Notes This release corresponds to the submitted manuscript version.