Network KPI Trajectory Sketches for Interpretability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The lack of interpretability in machine learning-based network assurance systems hinders the design of closed-loop control systems for predicting network key performance indicator (KPI) patterns, making it difficult to recognize subtle KPI patterns that may indicate network issues.
Innovation Solution
A service divides network KPI time series into chunks, clusters them, identifies representative sketches, associates labels with these sketches, and applies these labels to new KPI time series for pattern recognition and remediation actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning techniques are used to predict network KPI patterns, then prediction capability is improved, but interpretability deteriorates
Solution Approach 1:
The patent introduces trajectory sketches as intermediary representations that bridge the gap between complex machine learning predictions and human-understandable network patterns. These sketches extract and visualize key trajectory characteristics from KPI time series data, serving as a mediator that translates opaque ML outputs into interpretable visual forms while preserving prediction capability.
2Difficulty of detecting and measuring
If complex machine learning models are deployed for network assurance, then problem detection capability is improved, but system complexity increases
Solution Approach 1:
The patent extracts essential trajectory features from complex KPI time series data and represents them as simplified sketches. This extraction process isolates the critical pattern recognition functionality from the complex ML model, presenting only the necessary visual information for problem detection while reducing the cognitive complexity of the overall system.
3Measurement precision
If detailed KPI time series analysis is performed, then pattern recognition accuracy is improved, but processing time increases
Solution Approach 1:
The patent creates simplified copy representations (sketches) of the original KPI time series data that capture essential trajectory patterns. These sketch copies retain the critical information needed for accurate pattern recognition while requiring significantly less processing time and computational resources than analyzing the complete detailed time series data.
Data Source
AI summary
In one embodiment, a service divides one or more time series for a network key performance (KPI) into a plurality of time series chunks. The service clusters the plurality of time series chunks into a plurality of clusters. The service identifies a sketch that represents a particular one of the clusters. The service associates a label with the identified sketch. The service applies the label to a new KPI time series by matching the sketch to the new KPI time series.