Poster: A Deep-Learning Approach for ECG-Based Cryptographic Key Generation

This study proposes an innovative deep-learning method for generating biometric keys directly from individual ECG segments to improve secure authentication in cybersecurity systems. The model combines a convolutional neural network (CNN) with transformer-based attention to extract local and global features from fixed-length ECG records. Ground-truth keys are generated using a safe random method, acting as the standard for assessing key consistency and uniqueness. Initial findings suggest possible uses in safe, ECG-based biometric authentication systems.