Model Control Platform Stack for Probe-Based Failure Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current computer modeling systems face challenges in effectively monitoring and managing models, particularly in detecting performance issues before they become critical problems, and in rapidly deploying and remedying model performance failures in mission-critical scenarios.

Innovation Solution

A model control platform stack that includes a probe inventory and model inventory, allowing for the selection and deployment of probes and models to monitor specific locations, extract data, and adjust models based on inference, enabling early warning systems and rapid remediation of performance failures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional model monitoring methods are used, then model performance can be tracked, but performance issues cannot be detected early enough before becoming critical problems

Engineering Contradiction:
Improvemodel performance reliabilityVSAvoidtime to detect performance issues
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent implements preliminary action by deploying probes and monitoring agents before model failures occur. The system continuously collects performance metrics, logs, and system states in advance, enabling early detection of degradation trends. The proactive monitoring architecture captures data at multiple levels (model output, intermediate layers, system resources) before critical failures happen, allowing preventive interventions.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent establishes continuous feedback loops where model performance metrics are constantly monitored, analyzed, and fed back to the control system. The feedback mechanism compares actual performance against expected thresholds and triggers alerts or automated remediation when deviations are detected. This closed-loop feedback enables real-time performance tracking and rapid response to issues.

Inventive Principle:
Principle #23Feedback

2Reliability

If comprehensive model monitoring is implemented, then performance issues can be detected, but system complexity increases

Engineering Contradiction:
Improvemodel performance monitoringVSAvoidmonitoring system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the monitoring system into distinct modular components: probes for data collection, agents for local processing, central monitoring servers for analysis, and remediation modules for corrective actions. Each component has a specific function and can be independently deployed, configured, and maintained. This modular architecture reduces overall system complexity while enabling comprehensive monitoring coverage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces intermediary elements such as monitoring agents that act as mediators between the model system and the central monitoring infrastructure. These agents collect and pre-process data locally, filtering and aggregating information before transmitting to central servers. This intermediary layer simplifies the communication burden and reduces the complexity of direct point-to-point monitoring connections.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Stability of the object's composition

If model performance failures are not rapidly remediated, then system stability deteriorates, but rapid remediation requires complex deployment mechanisms

Engineering Contradiction:
Improvemodel system stabilityVSAvoidremediation deployment complexity
Core Design Contradiction:
Stability of the object's compositionVSDevice complexity

Solution Approach 1:

The patent implements preliminary action for remediation by pre-configuring backup models, rollback mechanisms, and recovery procedures before failures occur. The system maintains ready-to-deploy backup model versions and pre-established deployment pipelines that can be activated immediately when failures are detected. This advance preparation enables rapid remediation without requiring complex real-time decision-making or ad-hoc deployment procedures.

Inventive Principle:
Principle #10Preliminary action

4Reliability

If continuous model execution and monitoring is performed, then performance issues are detected early, but computational resources are consumed

Engineering Contradiction:
Improvecontinuous model monitoringVSAvoidcomputational resource consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent applies partial action by implementing selective monitoring strategies that focus computational resources on critical model components and high-risk operations. Instead of uniformly monitoring all model activities at maximum intensity, the system adjusts monitoring depth and frequency based on model importance, data sensitivity, and performance criticality. This selective approach maintains reliable monitoring while reducing overall computational overhead.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3792769A1Model control platform
Publication Date: 2021.03.17 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP3792769A1 patent drawingFigure 1
  • EP3792769A1 patent drawingFigure 2
  • EP3792769A1 patent drawingFigure 3

AI summary

The present disclosure relates to a system and a method for model control platform stack. The method includes, at an input layer of a model control platform stack, receiving input data. At a governance layer of the model control platform stack, the method includes maintaining a probe and model inventories; selecting a model, a monitoring location point, and a probe; and deploying, based on the selections of the probe and the model, a container to an orchestration layer of the model control platform stack. At the orchestration layer of the model control platform stack, the method includes accessing the container; using the container to deploy the probe and the model; scheduling an execution of the model to determine inference associated with the input data; during the execution, extracting probe data, using the probe, from the monitoring location point; and adjusting, based on the probe data and the inference, the model.