Hierarchical Multi-Model Generation for Cloud Site Forecasting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current techniques for monitoring and forecasting in cloud and network computing environments face challenges when a new site is added, as they consume significant computing resources and require extensive data gathering due to limited availability of raw telemetry data and key performance indicators, making it difficult to determine functionality, detect anomalies, and generate forecasts.

Innovation Solution

A hierarchical multi-model generation system that calculates a similarity score matrix from site data, groups it into clusters, identifies training and validation data, generates a meta model, and creates site-specific models to optimize forecasting and anomaly detection for new sites, using a telemetry-enhanced model agnostic meta learning approach.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional monitoring and forecasting techniques are used for new sites, then comprehensive data analysis can be performed, but computing resources are consumed significantly and extensive data gathering is required

Engineering Contradiction:
Improveforecasting accuracyVSAvoidcomputing resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies preliminary action by pre-training a meta-model on aggregated data from multiple sites before deployment to new sites. This pre-computed knowledge base enables new sites to leverage existing patterns without requiring extensive data gathering and model training from scratch, thereby reducing computing resource consumption while maintaining forecasting accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements universality through a meta-model that serves multiple sites simultaneously. The meta-model is trained on aggregated data from multiple sites and can be adapted to generate site-specific models for different locations, eliminating the need to train separate models for each site and reducing overall computing resource requirements.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If traditional monitoring techniques are used for new sites, then thorough anomaly detection can be achieved, but extensive data gathering is required due to limited availability of raw telemetry data

Engineering Contradiction:
Improveanomaly detection capabilityVSAvoiddata quantity required
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent applies preliminary action by pre-processing and aggregating data from multiple sites to create a comprehensive training dataset before model deployment. This preliminary data aggregation enables the meta-model to learn robust patterns that improve anomaly detection reliability at new sites without requiring extensive data gathering after deployment.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements merging by combining data from multiple sites to create an aggregated training dataset. This combined data pool enriches the training information available to the meta-model, enabling it to detect anomalies more reliably at new sites even when local data availability is limited.

Inventive Principle:
Principle #5Merging (Combining)

3Measurement precision

If site-specific models are trained from scratch for each new site, then accurate predictions can be generated, but training time is extended

Engineering Contradiction:
Improveprediction accuracyVSAvoidmodel training time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-training a meta-model on aggregated data from multiple sites before deployment. This pre-computed knowledge base enables new sites to leverage existing patterns without requiring extensive data gathering and model training from scratch, thereby reducing computing resource consumption while maintaining forecasting accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements copying by using the pre-trained meta-model as a template that can be quickly adapted to generate site-specific models. Instead of training models from scratch for each site, the system copies the learned patterns from the meta-model and fine-tunes them for specific sites, dramatically reducing training time while preserving prediction accuracy.

Inventive Principle:
Principle #26Copying

4Productivity

If comprehensive data gathering is performed for new sites, then model training can be improved, but computing resources are consumed significantly

Engineering Contradiction:
Improvemodel training efficiencyVSAvoidcomputing resource consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent implements universality through a meta-model that serves multiple sites simultaneously. The meta-model is trained on aggregated data from multiple sites and can be adapted to generate site-specific models for different locations, eliminating the need to train separate models for each site and reducing overall computing resource requirements.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent applies preliminary action by pre-training a meta-model on aggregated data from multiple sites before deployment to new sites. This pre-computed knowledge base enables new sites to leverage existing patterns without requiring extensive data gathering and model training from scratch, thereby reducing computing resource consumption while maintaining forecasting accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4280087A1Hierarchical multi-model generation in network and cloud computing systems
Publication Date: 2023.11.22 JUNIPER NETWORKS INC
  • EP4280087A1 patent drawingFigure 1A
  • EP4280087A1 patent drawingFigure 1B
  • EP4280087A1 patent drawingFigure 1C

AI summary

A device may receive site data identifying raw data or key performance indicators associated with a plurality of sites, and may calculate a similarity score matrix based on the site data. The device may group the site data into data clusters based on the similarity score matrix, and may identify training data and validation data based on the data clusters. The device may generate a meta model, and may train the meta model based on the training data. The device may validate the meta model based on the validation data, and may create site-specific models, for each of the plurality of sites, based on the meta model and the site data. The device may utilize the site-specific models with corresponding new site data of the plurality of sites to generate predictions for the plurality of sites.