Keystroke Biometric Verification Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing behavioral biometric methods rely on manually engineered features, which are fragile and complex, failing to be robust in practice, especially in uncontrolled environments, and have not fully leveraged the potential of deep learning for improved verification accuracy.
Innovation Solution
An automatic feature extraction framework using deep neural networks to generate determinate vectors for keystroke and gait-based verification, which automatically extracts latent features from high-dimensional spaces, providing a more robust and effective approach compared to traditional handcrafted features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manually engineered features are used for behavioral biometric verification, then domain-specific expertise can be leveraged, but the system becomes fragile and overly complex in practice
Solution Approach 1:
The system employs deep neural networks that automatically learn and extract behavioral biometric features from raw interaction data without requiring manual feature engineering. The neural network self-adjusts its feature extraction capabilities through training on labeled data, eliminating the need for domain experts to handcraft features while improving reliability through automated learning
Solution Approach 2:
The patent replaces the mechanical process of manual feature engineering with an automated neural network-based feature learning system. Instead of manually designing and adjusting feature extraction rules, the system uses deep learning algorithms to automatically discover relevant features from data, reducing complexity and improving adaptability
2Reliability
If manually engineered features are used, then initial performance in controlled experiments may be acceptable, but robustness fails in uncontrolled real-world environments
Solution Approach 1:
The system uses dynamic neural network models that can adapt their feature extraction strategies based on the input data characteristics. The network learns to handle variations in user behavior, device types, and environmental conditions through continuous training on diverse datasets, enabling robust performance in uncontrolled real-world settings
Solution Approach 2:
The patent employs neural networks that automatically adjust their internal parameters (weights and biases) based on training data from various environments. This parameter adaptation allows the system to generalize from controlled experimental conditions to uncontrolled real-world scenarios, improving robustness across different contexts
3Measurement precision
If traditional verification methods are used, then implementation is straightforward, but equal error rate remains high
Solution Approach 1:
The system uses deep neural networks that automatically learn optimal verification thresholds and decision boundaries from training data. The network self-calibrates its classification criteria through exposure to labeled examples of legitimate and fraudulent behavior, achieving lower equal error rates without manual threshold tuning
4Measurement precision
If more data is collected for training, then verification accuracy may improve, but data requirements and processing burden increase
Solution Approach 1:
The patent replaces traditional statistical methods that require large amounts of training data with deep neural networks that can learn effective representations from smaller datasets. The network's ability to automatically extract hierarchical features reduces the data hunger typical of traditional machine learning approaches while maintaining or improving verification accuracy
Data Source
AI summary
Behavioral verification of user identity includes building a deep neural network for keystroke-based behavioral verification of user identity. The building includes receiving recorded keystroke events, each such recorded keystroke event including (i) an indication of whether the recorded keystroke event is a key press or a key release, (ii) a key identifier of the respective key pressed or released, and (iii) a timestamp of the recorded keystroke event. The building further includes performing pre-processing of the recorded keystroke events to provide data structures representing sequential key events for processing by a deep neural network to extract local patterns, and training the deep neural network using the data structures. The method also includes providing the trained deep neural network for keystroke-based behavioral verification of user identity based on determinate vectors output from the trained deep neural network.