BrainNet: Improving Brainwave-based Biometric Recognition with Siamese Networks

With the advent of consumer wearables that capture brain activity, the use of brainwaves to verify a user's identity has been proposed as a convenient alternative to passwords. While recent work on brain biometrics shows feasible performance, it falls short in considering practical applicability. We propose a new solution, BrainNet, which trains a Siamese Network to measure the similarity of two electroencephalogram (EEG) inputs, and uses time-locked brain reactions instead of continuous mental activity to improve accuracy. This approach removes the need for retraining the brainwave recognition system, a common pitfall in current solutions, facilitating practical deployment. Furthermore, BrainNet achieves Equal Error Rates (EERs) of 0.14% in verification mode and 0.34% in identification mode, outperforming the state of the art even when evaluated under unseen attacker scenarios.