Multi-Stream RNN for System Metric Prediction Using Causal Attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional system metrics prediction approaches neglect the influence of system events, leading to incomplete and unreliable predictions, as they solely rely on historical metrics data without considering the causal relationships between events and metrics.
Innovation Solution
A multi-stream recurrent neural network (RNN) is employed to concurrently model multivariate metric series and event series, utilizing an attention mechanism to identify causality relations and predict future metrics, thereby improving prediction accuracy and providing explainable insights by ranking influential events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional approaches solely rely on historical metrics data for prediction, then the prediction process is simple, but the prediction accuracy and reliability are insufficient
Solution Approach 1:
The patent merges multiple data streams (multivariate metric series and event series) into a unified prediction model. The multi-stream RNN simultaneously processes both metric data and event log data, allowing the system to leverage complementary information from different sources to improve prediction accuracy while maintaining a coherent architectural framework.
Solution Approach 2:
The patent segments the prediction model into distinct functional components: separate RNN streams for processing metric series and event series, an attention mechanism for identifying influential events, and a prediction module. This segmentation allows each component to specialize in specific tasks while working together as an integrated system.
2Reliability
If event logs are incorporated into the prediction model, then the prediction reliability improves, but the computational complexity increases
Solution Approach 1:
The attention mechanism serves as an intermediary between the event series processing and the final prediction. It selectively identifies and weights influential events, filtering out noise and focusing computational resources on the most relevant event-metric relationships. This intermediary layer improves reliability by emphasizing causally significant events while managing computational complexity through selective processing.
Solution Approach 2:
The model dynamically adjusts its processing focus through the attention mechanism, which adaptively weights different events based on their relevance to metric fluctuations. Rather than treating all events uniformly, the system dynamically identifies and prioritizes influential events, allowing flexible adaptation to varying system conditions and event importance.
3Loss of information
If causality relations between events and metrics are modeled, then the interpretability of predictions improves, but the computational requirements increase
Solution Approach 1:
The attention mechanism provides feedback by identifying and highlighting the specific events that most influenced metric changes. This feedback loop allows the system to trace predictions back to their causal event sources, improving interpretability. The attention weights serve as explanatory signals that indicate which events drove particular prediction outcomes.
Solution Approach 2:
The patent creates a composite modeling approach that integrates multiple data types (metrics and events) and multiple processing mechanisms (RNN streams and attention) into a unified framework. This composite structure captures complex event-metric relationships while maintaining computational tractability through the coordinated interaction of its components.
Data Source
AI summary
A method for system metric prediction and influential events identification by concurrently employing metric logs and event logs is presented. The method includes concurrently modeling multivariate metric series and individual events in event series by a multi-stream recurrent neural network (RNN) to improve prediction of future metrics, where the multi-stream RNN includes a series of RNNs, one RNN for each metric and one RNN for each event sequence and modeling causality relations between the multivariate metric series and the individual events in the event series by employing an attention mechanism to identify target events most responsible for fluctuations of one or more target metrics.