AI Incident Prioritization With Feedback-Tuned Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cyber security systems face challenges in intelligently prioritizing large and complex security incidents across multiple computer systems, requiring uniform AI models while accommodating system-specific differences in importance, relevance, urgency, and severity.
Innovation Solution
An iterative process using pre-trained AI models that adapt feature extraction based on system-specific user feedback, optimizing prioritization for each computer system by adjusting feature precision without system-specific training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If uniform pre-trained AI models are used across all computer systems, then model consistency and ease of deployment are improved, but system-specific prioritization accuracy deteriorates
Solution Approach 1:
The patent applies local quality by allowing each computer system to have system-specific feature extraction parameters while using uniform pre-trained AI models. The feature extraction process is customized for each system based on user feedback, enabling local adaptation without requiring system-specific model training. This resolves the contradiction by maintaining model consistency at the AI model level while achieving system-specific accuracy through localized feature extraction adjustments.
2Adaptability or versatility
If system-specific feature extraction is implemented, then prioritization relevance is improved, but system complexity increases
Solution Approach 1:
The patent implements dynamics by making the feature extraction process adaptive and iterative. User feedback on prioritization results is continuously incorporated to refine feature extraction parameters for each system. This dynamic adjustment allows the system to become increasingly relevant to specific systems over time without requiring complex static configurations, resolving the contradiction between adaptability and complexity.
Solution Approach 2:
The patent applies feedback by using user responses to prioritization results to continuously improve system-specific feature extraction. User feedback loops allow the system to learn and adapt to each computer system's unique requirements, enhancing prioritization relevance while keeping the underlying AI models uniform and manageable, thus resolving the complexity-relevance trade-off.
3Productivity
If user feedback is aggregated across multiple computer systems, then overall model improvement is accelerated, but system-specific prioritization may be distorted
Solution Approach 1:
The patent applies segmentation by separating the learning process into two distinct layers: (1) uniform pre-trained AI models that benefit from aggregated cross-system learning, and (2) system-specific feature extraction parameters that are customized for each individual system. This segmentation allows productive aggregation of feedback at the model level while preserving system-specific accuracy through localized parameter adjustments, resolving the contradiction between improvement speed and system-specific precision.
Data Source
AI summary
A computer-implemented method for prioritizing security incidents includes maintaining one or more pre-trained Artificial Intelligence (AI) models that, when applied to features extracted from computer-system security incidents, produce a prioritization among the security incidents. An iterative process is run for each computer system among a plurality of computer systems. The iterative process includes detecting one or more security incidents occurring in the computer system, extracting features from the detected incidents, applying the one or more pre-trained AI models to the extracted features so as to produce a computer-system-specific prioritization, presenting the computer-system-specific prioritization to a user of the computer system, receiving user feedback that is indicative of a quality of the computer-system-specific prioritization, as decided by the user of the computer system, and adjusting subsequent extraction of the features, for the computer system, based on the user feedback.