Multiuser Learning System for Detecting Diverse Rare Behaviors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting rare events in time-series data from financial and industrial applications often fail to cover a diverse range of behaviors, leading to missed important types of criminal or system failures due to limited investigation resources.
Innovation Solution
A computer-implemented method that processes time-series data from multiple sources, generates entity profiles, extracts features, compares them to identify differences, trains models, and updates them based on feedback to efficiently detect and prioritize rare behaviors, incorporating latent features and diversity metrics to ensure a wide variety of outlier behaviors are investigated.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a relatively large team of investigators is deployed to detect rare events, then the detection coverage increases, but the time and resource consumption increases proportionally
Solution Approach 1:
The patent segments the investigation process into two distinct phases: (1) an automated AI-driven screening phase that processes all entities to identify potential rare events, and (2) a human investigator phase that focuses only on the filtered subset of high-priority cases. This segmentation allows the system to maintain high detection coverage while dramatically reducing the time and resources required for human investigation.
Solution Approach 2:
The patent introduces an AI model as an intermediary between the raw data and human investigators. This intermediary automatically analyzes time-series data, extracts features, and prioritizes entities based on their likelihood of exhibiting rare behaviors. The AI intermediary filters and ranks entities before human review, enabling investigators to focus their expertise on the most critical cases rather than manually examining all entities.
2Adaptability or versatility
If traditional detection methods are used, then the investigation process is simple, but diverse rare behaviors are missed due to limited resource exploration
Solution Approach 1:
The patent transforms the detection approach by changing key parameters: instead of using simple rule-based thresholds, it employs machine learning models that analyze multiple features simultaneously. The system extracts diverse features from time-series data including statistical measures, pattern recognition metrics, and behavioral characteristics. These parameter changes enable the system to detect a broader variety of rare behaviors while managing complexity through automated feature extraction and model-based analysis.
Solution Approach 2:
The patent performs preliminary actions by pre-processing time-series data and pre-extracting features before the actual detection process. The system pre-computes statistical features, patterns, and behavioral metrics for all entities, storing them in an optimized format. This preliminary processing enables the detection models to quickly analyze diverse behaviors without requiring complex real-time computations during the investigation phase.
3Reliability
If manual investigation of all entities is performed, then comprehensive coverage is achieved, but the cost of time and resources becomes prohibitive
Solution Approach 1:
The patent implements a feedback mechanism where the AI model is trained on labeled data from investigator decisions. When investigators confirm or reject detected rare events, this feedback is used to retrain and refine the AI model, improving its accuracy over time. The system continuously learns from investigator feedback, adjusting its detection thresholds and feature weights to better identify true rare events while reducing false positives, thereby maintaining high detection completeness with improved efficiency.
Solution Approach 2:
The patent employs dynamic prioritization where the detection thresholds and model parameters are adjusted based on the characteristics of the data being analyzed. The system dynamically adapts its sensitivity levels, feature importance weights, and ranking criteria based on the specific time period, entity types, and observed behavior patterns. This dynamic approach allows the system to maintain high detection completeness across varying conditions while optimizing resource allocation for each investigation batch.
Data Source
AI summary
A method, a system, and a computer program product for detecting a diverse set of rare behavior. A time-series data representing one or more actions executed by an entity is received from a plurality of time-series data sources and is processed. A data structure corresponding to the entity, identifying the entity, and including one or more representations of processed time-series data identifying the actions is generated. A current action executed by the entity is detected. Current time-series data corresponding to the current action is received and associated with the data structure. First features are extracted from the generated data structure based on current time-series data and compared to second features extracted for at least another entity to determine difference parameters between first and second features. One or more models are trained using difference parameters, and a score for each action executed by the entity is determined. An action is identified based on the determined scores and the training of the models is updated in response to receiving a feedback data to the identified action, and at least another action is identified. A consistency score is generated for the feedback data.