Curve Segmentation Retraining Using Frequency and Spatial Drift Scores

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing workflows utilizing deep learning models face challenges in identifying when to fine-tune or retrain models due to data distribution shifts and data privacy issues, particularly in workflows with multiple machine learning models.

Innovation Solution

A curve retraining decision system that uses frequency and spatial models to generate scores for a curve segmentation model of a raster digitization engine, automatically determining when to retrain the model based on these scores.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual monitoring and retraining of machine learning models is performed, then model accuracy can be maintained, but computational resources and time are wasted due to frequent unnecessary retraining

Engineering Contradiction:
Improvemodel accuracyVSAvoidretraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system implements automatic feedback mechanisms by continuously monitoring data distribution shifts and model performance metrics. When the drift detector identifies significant distribution changes or when performance drops below thresholds, the system automatically triggers retraining workflows, eliminating the need for manual monitoring while optimizing retraining timing to avoid unnecessary computational waste.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The machine learning workflow system performs self-diagnosis and self-retraining by automatically detecting data distribution shifts and performance degradation. The system monitors its own models, determines when retraining is necessary based on drift metrics, and executes retraining workflows autonomously, reducing both manual intervention and unnecessary retraining operations.

Inventive Principle:
Principle #25Self-service

2Reliability

If frequent retraining is performed to maintain model accuracy, then model performance improves, but computational resources are excessively consumed

Engineering Contradiction:
Improvemodel accuracyVSAvoidcomputational resources
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system uses feedback from drift detection metrics and performance monitoring to intelligently control retraining frequency. By continuously measuring data distribution changes and model performance, the system only initiates retraining when necessary, avoiding wasteful computational resource consumption while maintaining model accuracy through targeted retraining events.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system dynamically adjusts retraining parameters based on detected drift severity and performance degradation levels. When minor drift is detected, the system may apply lighter retraining interventions; when significant drift occurs, full retraining is triggered. This parameter-based control optimizes computational resource usage while maintaining model effectiveness.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If data is shared across multiple workflows for model training, then model performance improves, but data privacy and residency issues arise

Engineering Contradiction:
Improvemodel performanceVSAvoiddata privacy issues
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The system segments data handling by workflow, allowing each machine learning workflow to access and train on its own localized data while maintaining model performance through architecture-level optimizations. This segmentation approach enables models to achieve high accuracy on their specific tasks without requiring access to other workflows' sensitive data, thereby resolving privacy concerns.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Each workflow is configured with local data access permissions and training data quality is optimized for its specific purpose. The system allows data to be processed locally within each workflow context with appropriate quality controls, eliminating the need to share data across workflows while maintaining high model performance through localized optimization.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4704052A1Curve retraining decision system
Publication Date: 2026.03.04 SERVICES PETROLIERS SCHLUMBERGER SA
  • EP4704052A1 patent drawingFigure 1
  • EP4704052A1 patent drawingFigure 2
  • EP4704052A1 patent drawingFigure 3

AI summary

A method implements a curve retraining decision system. The method includes executing a frequency model using an initial image and an extracted curve image to generate a frequency model score. The method further includes executing a spatial model using the initial image and the extracted curve image to generate a spatial model score. The method further includes generating a curve retraining score for a curve segmentation model of a raster digitization engine. The method further includes retraining the curve segmentation model using the curve retraining score.