Time-Series Outlier Detection Using Unsupervised GAN Windowing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional performance monitoring and anomaly detection in networks are manual, expensive, time-consuming, and require human expertise, failing to scale with complex networks and lacking accuracy in detecting multi-dimensional failures.
Innovation Solution
An unsupervised method using Deep Neural Networks (DNNs) with windowing techniques to process time-series data, enabling pattern detection and localization of anomalies without human intervention, employing GANs and BiGANs for improved performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual performance monitoring and anomaly detection are used, then human expertise can interpret complex failures, but the process is expensive, time-consuming, and does not scale with large networks
Solution Approach 1:
The patent replaces manual mechanical inspection and human expert analysis with an automated image processing system using convolutional neural networks. The system automatically captures performance monitor screenshots, processes them through deep learning models, and detects anomalies without human intervention, thereby reducing reaction time while maintaining or improving detection accuracy.
Solution Approach 2:
The system enables self-service anomaly detection by automatically monitoring performance metrics, generating visual representations, and identifying issues without requiring human expertise. The automated pipeline includes data collection, image generation, neural network processing, and alert generation, allowing the system to serve itself in detecting and reporting anomalies.
2Adaptability or versatility
If rule-based engines with hard-coded thresholds are used for anomaly detection, then known failures can be detected, but the approach cannot find failures spanning multiple network elements and requires extensive engineering time
Solution Approach 1:
The patent transforms traditional one-dimensional threshold-based detection into a two-dimensional image-based approach. By representing performance metrics as visual images with multiple dimensions (time, metrics, network elements), the convolutional neural network can detect complex patterns and correlations across multiple network elements simultaneously, overcoming the limitations of simple threshold rules.
Solution Approach 2:
The image processing system serves multiple functions: it visualizes performance data, detects anomalies, identifies patterns across network elements, and generates alerts. This universal approach replaces multiple specialized rule-based engines with a single flexible system that can adapt to different failure types and network configurations.
3Ease of manufacture
If simple one-dimensional thresholding rules are used, then implementation is straightforward, but accuracy is limited and cannot detect complex multi-dimensional failure patterns
Solution Approach 1:
The patent replaces simple mechanical threshold comparison with sophisticated image processing using convolutional neural networks. The system automatically transforms performance data into visual images and uses deep learning to detect complex patterns, achieving high accuracy while maintaining automated implementation without requiring expert tuning of threshold rules.
Data Source
AI summary
Systems and methods for detecting patterns in data from a time-series and for detecting outliers in network data in an unsupervised manner are provided. In one implementation, a method includes the steps of obtaining network data from a network to be monitored and creating a window from the obtained network data. The method also includes the step of detecting outliers of the obtained data with respect to the window using an unsupervised deep learning process (e.g., using a Generalized Adversarial Network (GAN) learning technique and/or a Bidirectional GAN (BiGAN) learning technique) for enabling the learning of a data distribution. The unsupervised process, for example, does not require manual intervention.