ML Task Scheduling for Lower-Carbon Model Training

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The increasing computational demands of machine learning tasks result in significant carbon dioxide emissions due to energy-intensive computing, necessitating a system that reduces emissions while maintaining performance.

Innovation Solution

A computer system that schedules machine learning tasks across multiple data processing devices, selecting algorithms, devices, and execution times to optimize renewable energy usage and minimize carbon dioxide emissions by considering site-specific power sources and availability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If computational resources are increased to meet growing machine learning task demands, then model training performance and accuracy are improved, but carbon dioxide emissions and energy consumption increase significantly

Engineering Contradiction:

Improvemodel training accuracyVSAvoidcarbon dioxide emissions

Core Design Contradiction:

Manufacturing precisionVSObject-generated harmful factors

Solution Approach 1:

The patent combines multiple computational tasks that can be executed in parallel across available computing resources, optimizing the utilization of existing infrastructure rather than continuously adding new resources. This merging approach maintains training accuracy while reducing the carbon footprint by efficiently using available clean energy capacity.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system dynamically adjusts computational task scheduling based on real-time or predicted carbon intensity of energy sources. By shifting task execution timing to periods when renewable energy is abundant, the system maintains model training performance while adapting to varying energy conditions, thereby reducing carbon dioxide emissions without sacrificing accuracy.

Inventive Principle:

Principle #15Dynamics

2Object-generated harmful factors

If computational tasks are executed during periods of high renewable energy availability, then carbon dioxide emissions are reduced, but task execution timing and scheduling complexity increase

Engineering Contradiction:

Improvecarbon dioxide emissionsVSAvoidscheduling system complexity

Core Design Contradiction:

Object-generated harmful factorsVSDevice complexity

Solution Approach 1:

The system performs preliminary assessment of energy source carbon intensity and predicts optimal execution windows for computational tasks. By pre-identifying periods of high renewable energy availability and preparing task schedules in advance, the system reduces emissions without requiring complex real-time scheduling decisions, thus managing complexity effectively.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary scheduling layer that sits between the computational tasks and the physical computing resources. This intermediary component abstracts the complexity of carbon intensity monitoring and task timing optimization, managing the transition from energy data to execution schedules while reducing overall system complexity through modular design.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Measurement precision

If machine learning model size and training data size are increased, then model performance and accuracy are improved, but energy consumption and computational resource requirements increase

Engineering Contradiction:

Improvemodel accuracyVSAvoidenergy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system changes the temporal parameter of task execution by scheduling large-scale model training during periods when renewable energy is abundant. By altering when computational intensive tasks are performed rather than changing the models themselves, the system maintains high model accuracy while reducing energy consumption from carbon-intensive sources.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system discards the constraint of immediate task execution and recovers computational capacity during periods of high renewable energy availability. By being willing to delay task execution and recover computing power when clean energy is abundant, the system can handle larger models and datasets without proportionally increasing carbon emissions.

Inventive Principle:

Principle #34Discarding and recovering

Data Source

PatentUS12444942B2Systems and methods for reducing carbon emissions from machine learning computational tasks

Publication Date: 2025.10.14 GE INFRASTRUCTURE TECH LLC

AI summary

A computer system is provided. The computer system includes a scheduling computing device configured to receive computational task data defining a computational task to be performed, retrieve site data corresponding to each of a plurality of data processing computing devices, select, based on the computational task data and the site data, i) a first computational algorithm for executing the computational task, ii) a first data processing computing device of the plurality of data processing computing devices, and iii) at least one time period for executing the first computational algorithm by the first data processing computing device, wherein the first computational algorithm, the first data processing computing device, and the at least one time period are selected to facilitate reducing carbon dioxide emissions associated with executing the computational algorithm, and instruct the first data processing computing device to execute the first computational algorithm during the at least one time period.