Training Data Generation via Application Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating user operation models for new applications are inefficient, requiring extensive time and work to acquire and update operation histories, especially when applications are updated or installed on real terminals.
Innovation Solution
A training data generating system that identifies existing applications with similar features to a target application, calculates the degree of similarity, and generates training data for the target application from the similar existing applications, allowing for efficient construction of user operation models without the need for new operation histories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If operation history is acquired from real terminal and model is created each time application is updated or installed, then the user operation model corresponds to the latest application version, but the time and work required becomes too extensive
Solution Approach 1:
The system performs preliminary actions by acquiring operation history and creating user operation models in advance for existing applications before they are updated or replaced. When a new application version is installed, the system can leverage these pre-created models and their associated tag information to rapidly generate updated models without restarting the entire data collection and model creation process from scratch.
Solution Approach 2:
The system creates copies of existing user operation models and their tag information when applications are updated. Instead of creating entirely new models, the system copies the structure and learned patterns from existing models and adapts them to the new application version, significantly reducing model creation time while maintaining accuracy.
2Quantity of substance
If manual creation or recording of normal logs is performed, then the logs can be prepared for cybersecurity exercises, but the process requires extended time and tedious work
Solution Approach 1:
The system enables self-service log generation by automatically creating normal logs through the user operation model without requiring manual creation or recording. The model autonomously generates logs that reflect typical user behavior patterns, eliminating the need for human operators to manually create or record logs for cybersecurity exercise preparation.
Solution Approach 2:
The system replaces the mechanical process of manual log creation and recording with an automated computational system. The user operation model, trained on historical operation data, automatically generates synthetic normal logs that capture realistic user behavior patterns, substituting human manual work with algorithmic generation.
3Productivity
If the same normal logs are reused for multiple cybersecurity exercises, then the work is reduced, but attendees cannot improve their investigative skills
Solution Approach 1:
The system introduces dynamics to the log generation process by enabling on-demand creation of new normal logs through the user operation model. When cybersecurity exercises require fresh data, the system can dynamically generate new logs with varied characteristics while maintaining realistic user behavior patterns, ensuring that each exercise provides novel challenges for skill development.
Data Source
AI summary
A training data generating system includes an existing tag information acquiring unit, a target tag information acquiring unit, a selecting unit, and a training data generating unit. The existing tag information acquiring unit identifies, as an existing application, an application for which training data for a user operation model has already been generated based on a user operation history and, for each existing application, acquires tag information indicating a feature of the application as existing tag information. The target tag information acquiring unit acquires tag information of a target application as target tag information. The selecting unit calculates a degree of similarity between the target tag information and each piece of the existing tag information and, based on the degree of similarity, selects a similar application from the existing applications. The training data generating unit generates training data for the target application from training data for the similar application.