Event-Driven ML Agent Architecture for Easier Model Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Integrating machine learning algorithms in an existing operating environment is challenging due to difficulties in deployment and configuration.
Innovation Solution
An event-driven architecture is implemented with a first and second machine-learning agent, where the first agent generates events with metadata, publishes them in a virtualized space, and the second agent determines and processes these events for further training or prediction, enabling seamless integration and processing of machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning algorithms are integrated into an existing operating environment, then the system gains advanced processing capabilities, but the deployment and configuration complexity increases significantly
Solution Approach 1:
The patent introduces an event-driven architecture as an intermediary layer between the existing operating environment and machine learning algorithms. This architecture uses event producers, event consumers, and event streams to mediate interactions, thereby simplifying the integration process and reducing deployment complexity while maintaining advanced processing capabilities
Solution Approach 2:
The system segments machine learning operations into discrete events with specific types, producers, and consumers. By dividing the complex ML integration into modular event-based components, the system reduces configuration complexity while preserving functional capabilities
2Adaptability or versatility
If multiple machine learning agents are deployed to process different events, then the system's processing versatility improves, but the architecture complexity increases
Solution Approach 1:
The patent creates a universal event-driven framework that can accommodate multiple machine learning agents with different functions. The standardized event structure and processing mechanisms allow diverse ML agents to operate within a single unified architecture, increasing versatility without proportionally increasing complexity
Solution Approach 2:
The system employs dynamic event routing where consumers subscribe to event types they can handle. This dynamic configuration allows the architecture to adapt to different ML agent combinations without requiring complex static design, balancing versatility and complexity
3Measurement precision
If machine learning models are continuously trained and updated, then the system's predictive accuracy improves, but the computational resource consumption increases
Solution Approach 1:
The system implements periodic training triggers based on event accumulation or time intervals rather than continuous training. Event producers generate training events that trigger model updates at optimized intervals, improving predictive accuracy while reducing unnecessary computational resource consumption
Solution Approach 2:
The event-driven architecture incorporates feedback mechanisms where model performance is monitored through events, and training is triggered only when performance thresholds are not met or significant data patterns emerge. This feedback-based approach optimizes the balance between accuracy improvement and resource consumption
Data Source
AI summary
There is disclosed a method and system for operating an event-driven architecture. The event-driven architecture comprises a first machine-learning (ML) agent operating a first service and a second ML agent operating a second service. The first ML agent comprises a first model and first model metadata. The second ML agent comprises a second model and second model metadata. The method comprises generating, by the first ML agent, an event associated with event metadata. The event comprises results generated by the first model. The event metadata comprises an event identifier (ID). The first ML agent publishes the event in a virtualized dedicated space. The second ML agent determines whether the event is to be processed by the second ML agent. If a determination is made that the message is to be processed by the second ML agent, the second ML agent processes the event to generate an output.