Keystroke User Identification Using Synthetic Negative Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional keystroke biometric systems for user identification face challenges in real-world scenarios due to the lack of sufficient negative examples, leading to increased prediction errors, as they often rely on statistical principles that entangle the user identification task with text classification, making it impractical to collect negative keystroke samples from other users.
Innovation Solution
A generative partner-adversarial network (GPAN) is introduced to automatically generate synthetic negative keystroke samples, comprising a multi-modal generator, multi-class discriminator, and regression neural network, trained on user keystroke data to produce temporally coherent keystroke sequences for user identification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional statistical principles are used for user identification, then the system is simpler to implement, but the user identification accuracy deteriorates due to entanglement with text classification and lack of negative examples
Solution Approach 1:
The patent employs a generative adversarial network (GAN) to create synthetic negative keystroke samples that copy the statistical properties and temporal patterns of real keystroke data. The generator network learns to produce fake keystroke sequences from impostor users, which then serve as training data for the binary classifier, enabling accurate user identification without collecting actual negative examples from real users.
Solution Approach 2:
The patent introduces a discriminator network as an intermediary between the generator and the training process. The discriminator acts as a critic that evaluates the authenticity of generated keystroke samples, providing feedback to the generator to improve sample quality. This intermediary mechanism enables the system to iteratively refine synthetic negative examples, resolving the contradiction between implementation simplicity and identification accuracy.
2Measurement precision
If negative keystroke samples from other users are collected, then the binary classification accuracy is improved, but the data collection complexity and practical feasibility worsen
Solution Approach 1:
The system performs self-service by automatically generating its own negative training samples through the GAN architecture. Instead of requiring external data collection from multiple users, the generator network autonomously creates synthetic impostor keystroke samples by learning from available training data, eliminating the need for complex multi-user data collection processes while maintaining high classification accuracy.
Solution Approach 2:
The generator network copies the essential characteristics of real keystroke data—including temporal patterns, key press durations, and typing rhythms—to create realistic synthetic negative samples. This copying mechanism allows the system to obtain abundant negative training data without the practical difficulties of collecting genuine impostor keystrokes from real users.
3Measurement precision
If synthetic negative keystroke samples are generated using GAN, then the user identification accuracy is improved, but the system complexity and training time increase
Solution Approach 1:
The patent segments the complex GAN training process into distinct phases: first training the generator to produce realistic keystroke samples, then training the discriminator to evaluate them, and finally using the combined system to generate negative samples for binary classification training. This segmentation of the training process manages system complexity by breaking down the challenging GAN training into manageable stages, enabling the system to achieve high user identification accuracy.
4Ease of manufacture
If only positive keystroke samples from the target user are used, then the training process is simpler, but the ability to distinguish legitimate users from attackers deteriorates
Solution Approach 1:
The generator network copies the statistical properties and temporal dynamics of legitimate user keystrokes to create synthetic impostor samples. By learning the patterns of genuine typing behavior, the generator produces realistic fake samples that capture the essence of legitimate user behavior, enabling the binary classifier to learn the distinction between real and fake keystrokes without requiring actual impostor data.
Solution Approach 2:
The patent employs dynamic training where the generator and discriminator networks continuously adapt to each other. The generator dynamically adjusts its sample generation based on discriminator feedback, creating increasingly realistic negative samples. This dynamic interaction between networks enables the system to achieve high authentication reliability by learning from evolving synthetic negative examples rather than static positive samples alone.
Data Source
AI summary
An apparatus adapted to identify a user based on keystroke dynamics of an input by the user, the apparatus adapted to: execute a first training phase of training a keystroke sample generator to generate negative keystroke samples; execute a second training phase of training a user identification model based at least in part on a plurality of negative keystroke samples generated using the keystroke sample generator; and execute a deployment of the user identification model to authenticate an input sample associated with the user using the trained user identification model.