Deep Neural Network Keystroke Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing behavioral biometric methods rely on manually engineered features, which are fragile and overly complex, failing to be robust in practice, especially in realistic datasets, and have not been effectively applied to keystroke and gait verification with the success seen in other biometric domains like facial and speaker recognition.
Innovation Solution
An automatic feature extraction framework using deep neural networks to generate determinate vectors, which are used for keystroke and gait verification, providing a more robust and efficient method for user identification by learning latent features from high-dimensional spaces without domain-specific knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manually engineered features are used for behavioral biometric verification, then domain-specific expertise can be applied, but the features become fragile and overly complex, failing to be robust in practice
Solution Approach 1:
The patent replaces manual feature engineering with an automatic feature extraction system based on deep neural networks. The DNN automatically learns discriminative features from raw behavioral data (keystroke patterns, gait patterns) without requiring manual domain-specific feature design, thereby reducing complexity while improving robustness through data-driven learning
Solution Approach 2:
The system enables the data itself to speak for the user by allowing the deep neural network to automatically extract meaningful features directly from raw behavioral data. The feature extraction process is self-supervised, where the network learns to identify discriminative patterns without manual intervention, making the system more adaptive and robust to variations in real-world data
2Adaptability or versatility
If traditional behavioral biometric methods are applied to realistic datasets, then they can handle real-world variability, but performance (equal error rate) deteriorates compared to controlled experiments
Solution Approach 1:
The patent changes the fundamental parameters of feature extraction by transitioning from handcrafted features to deep learned features. The deep neural network learns optimal feature representations directly from realistic data distributions, adapting to real-world variability while maintaining high verification accuracy through hierarchical feature learning and data-driven parameter optimization
Solution Approach 2:
The system segments the feature extraction process into multiple hierarchical layers within the deep neural network. Each layer learns progressively more abstract and discriminative features from the raw behavioral data, enabling the system to capture both fine-grained details and broader patterns that contribute to robust verification performance on realistic datasets
3Extent of automation
If deep learning is applied to behavioral biometrics, then automatic feature extraction can be achieved, but the method has not yet been applied with the same success as in facial and speaker recognition
Solution Approach 1:
The patent successfully applies deep learning to behavioral biometrics by replacing manual feature engineering with automatic deep feature extraction. The deep neural network processes raw behavioral sequences (keystroke timing, gait patterns) and automatically learns discriminative representations, achieving verification accuracy comparable to established biometric systems while fully automating the feature extraction pipeline
Data Source
AI summary
A method for keystroke-based behavioral verification of user identity of a subject user of a computer system includes obtaining an enrollment signature corresponding to an identified user and serving as a unique identifier of the identified user, the enrollment signature including an enrollment determinate vector generated based on supplying enrollment keystroke data to a deep neural network for processing. The method further includes obtaining verification determinate vector(s), the verification determinate vector(s) for comparison to the enrollment signature to determine whether the subject user is the identified user. The method compares the verification determinate vector(s) to the enrollment signature and generates a probability indicator indicating a probability that keystroke data from a common user produced, from the deep neural network, the enrollment signature and the verification determinate vector(s), and indicates to the computer system whether, based on the probability indicator, the subject user is verified to be the identified user.