Transformer-based Deep Learning for Imagined Speech Decoding using KARA ONE Dataset

This work represents my Master’s thesis, which received a perfect score of 20/20.

Highlights:

• Utilized the KARA ONE dataset for multi-class imagined speech classification.
• Implemented a Transformer model to capture long-range dependencies in EEG time-series data.
• Demonstrated that self-attention mechanisms are superior to traditional convolutions for non-stationary neural signals.