🤖 AI Summary
This work addresses the challenge of cross-device (desktop/mobile) keystroke dynamics authentication for large-scale users under a few-shot setting—only five enrollment samples per user. To achieve high generalization, we propose three key innovations: (1) synthesizing temporal features to model individual typing behavioral biases; (2) a dual-branch RNN-CNN architecture incorporating both channel-wise and temporal attention mechanisms; and (3) a Set2Set loss function that preserves global embedding-space structure, coupled with a progressive hard-sample curriculum learning strategy. Evaluated on benchmarks comprising 15,000 and 5,000 users, our method achieves average equal-error rates (EER) of 0.77% and 1.03%, respectively, and single-threshold global EERs of 3.25% and 3.61%—substantially outperforming prior approaches. It constitutes the winning solution of the KVC-onGoing competition.
📝 Abstract
In 2021, the pioneering work TypeNet showed that keystroke dynamics verification could scale to hundreds of thousands of users with minimal performance degradation. Recently, the KVC-onGoing competition has provided an open and robust experimental protocol for evaluating keystroke dynamics verification systems of such scale. %, including considerations of algorithmic fairness. This article describes Type2Branch, the model and techniques that achieved the lowest error rates at the KVC-onGoing, in both desktop and mobile typing scenarios. The novelty aspects of the proposed Type2Branch include: i) synthesized timing features emphasizing user behavior deviation from the general population, ii) a dual-branch architecture combining recurrent and convolutional paths with various attention mechanisms, iii) a new loss function named Set2set that captures the global structure of the embedding space, and iv) a training curriculum of increasing difficulty. Considering five enrollment samples per subject of approximately 50 characters typed, the proposed Type2Branch achieves state-of-the-art performance with mean per-subject Equal Error Rates (EERs) of 0.77% and 1.03% on evaluation sets of respectively 15,000 and 5,000 subjects for desktop and mobile scenarios. With a fixed global threshold for all subjects, the EERs are respectively 3.25% and 3.61% for desktop and mobile scenarios, outperforming previous approaches by a significant margin. The source code for dataset generation, model, and training process is publicly available.