Containerized ML Retraining with Dynamic Resource Weighting

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Cloud-based machine learning model training and retraining face issues such as resource bottlenecks due to unsynchronized retraining schedules and inefficient resource allocation across multiple entities, leading to unnecessary resource usage and bottlenecks.

Innovation Solution

An applications intelligence framework with intelligent scheduling and dynamic weighted container assignment components, utilizing machine learned models to optimize retraining frequency and resource allocation based on data variation, user feedback, and algorithm type, ensuring efficient resource utilization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If retraining schedules are set at preset intervals for multiple entities, then model maintenance is simplified, but resource bottlenecks occur due to unsynchronized usage timing

Engineering Contradiction:

Improvemodel maintenance simplicityVSAvoidresource utilization efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The system dynamically adjusts retraining schedules based on real-time resource availability and entity-specific needs, transitioning from static preset intervals to adaptive timing. The cloud-based platform monitors resource usage patterns and automatically reschedules retraining operations to distribute load across available capacity, preventing bottlenecks while maintaining model freshness.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system implements feedback loops that monitor resource usage, model performance metrics, and data drift patterns to continuously optimize retraining schedules. This feedback mechanism allows the system to learn from past resource allocation patterns and adjust future scheduling decisions, balancing operational simplicity with resource efficiency.

Inventive Principle:

Principle #23Feedback

2Reliability

If retraining operations are performed frequently to ensure model accuracy, then model performance is improved, but unnecessary resource usage increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidcloud resource consumption

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The system changes key parameters including retraining frequency, data sampling rates, and model complexity based on actual performance needs and resource availability. By dynamically adjusting these parameters, the system performs retraining only when necessary to maintain accuracy, avoiding unnecessary resource consumption during periods of stable model performance.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system applies partial retraining actions by selectively updating only the portions of the model that require improvement based on data drift analysis, rather than performing complete retraining cycles. This partial action approach maintains model accuracy while significantly reducing computational resource requirements compared to full retraining.

Inventive Principle:

Principle #16Partial or excessive action

3Productivity

If cloud resources are shared among multiple entities, then resource utilization efficiency is improved, but bottlenecks occur due to concurrent retraining demands

Engineering Contradiction:

Improveresource utilization efficiencyVSAvoidmodel training reliability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system segments the shared cloud resource pool into isolated containers or virtual environments for different entities, allowing concurrent retraining operations to proceed independently without interference. This segmentation maintains high resource utilization while preventing bottlenecks by enabling parallel processing across multiple isolated training jobs.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system introduces an intermediary scheduling layer that manages resource allocation between multiple entities and the shared cloud infrastructure. This intermediary component mediates conflicting retraining demands by prioritizing jobs based on urgency, resource requirements, and available capacity, ensuring reliable model training while maximizing shared resource utilization.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260105384A1Dynamically scalable machine learning model generation and retraining through containerization

Publication Date: 2026.04.16 SAP SE

AI summary

In an example embodiment, a model generation component may additionally assign various cloud resources to a machine learned model so that the training or retraining of the model can be performed using these resource. The containers may be weighted to handle model generation work of different weight. Having one single configuration for a container responsible for generating all models leads to overuse of hardware resources because machine learning algorithms are very resource intensive, and thus dynamically selecting the weight improves hardware utilization.