Temporal Logic Composite Event Detection for Real-Time Decisions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning methods for composite event estimation in dynamic multivariate temporal data fail to provide a constructive structural and interpretable view, neglecting domain-specific knowledge and leading to poor performance in real-time decision support.
Innovation Solution
A computer-implemented method using temporal logic to identify temporally related atomic events from multivariate datasets, incorporating domain-specific knowledge and machine learning to discover composite events, with supervised or reinforcement learning to construct rules for accurate event detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning models are used for composite event estimation, then real-time decision support is enabled, but the structural interpretability and domain knowledge integration deteriorate
Solution Approach 1:
The patent segments composite events into atomic events with distinct predicates and temporal relations. Each atomic event is independently identified and annotated, then combined through temporal logic to form composite events. This segmentation enables both real-time processing of individual events and systematic composition of meaningful patterns, resolving the contradiction between speed and interpretability.
Solution Approach 2:
The patent introduces temporal logic as an intermediary framework between raw multivariate data and high-level composite event detection. Temporal logic rules serve as mediators that systematically combine atomic events with specific temporal relations (before, after, during, concurrently) to form composite events, providing both computational efficiency and logical interpretability.
2Speed
If conventional deep learning approaches are used, then processing speed is improved, but accuracy on benchmark composite event detection tasks deteriorates
Solution Approach 1:
The patent performs preliminary actions by pre-defining atomic event predicates and temporal logic rules before processing new data. The system pre-annotates training data with atomic event labels and temporal relations, then uses this pre-processed information to rapidly detect composite events in real-time, achieving both speed and accuracy.
Solution Approach 2:
The patent changes the fundamental parameters of event detection from continuous deep learning outputs to discrete atomic event predicates with explicit temporal relations. This parameter transformation enables exact matching against predefined composite event patterns, significantly improving detection accuracy while maintaining processing speed through efficient rule-based evaluation.
3Measurement precision
If domain-specific knowledge is integrated into the model, then event detection accuracy is improved, but model complexity and data requirements increase
Solution Approach 1:
The patent extracts domain-specific knowledge into separate, reusable temporal logic rules and atomic event predicates. Instead of embedding all domain knowledge within a single complex model, the system extracts and organizes knowledge into discrete components that can be independently managed, updated, and reasoned about, reducing overall system complexity.
Solution Approach 2:
The patent creates a universal temporal logic framework that can handle multiple types of composite events across different domains using the same underlying mechanisms. The same atomic event annotation and temporal logic rule evaluation process applies universally, reducing the need for domain-specific model variations and simplifying data processing requirements.
Data Source
AI summary
A computer-implemented method of discovering a composite durational event structure through temporal logic includes identifying a plurality of temporally related atomic events from temporal data trajectories of a multivariate dataset according to a definition of an atomic event predicate. At least one composite event having a durational event structure of at least some of the plurality of the temporally related atomic events is discovered by machine learning. An action is performed that is selected from a predetermined list associated with the composite event.