MLSUS01-BP01 Define the overall environmental impact or benefit - Machine Learning Lens
Implementation guidanceResources

MLSUS01-BP01 Define the overall environmental impact or benefit

Measure your workload's impact and its contribution to the overall sustainability goals of the organization. Understand the environmental footprint of machine learning systems to make informed decisions about resource allocation and optimization.

Desired outcome: You gain a clear understanding of how your ML workload impacts your organization's sustainability objectives, including energy consumption, carbon emissions, and resource utilization. You have established specific sustainability objectives and success criteria to measure against, which assists you when optimizing your ML workloads and demonstrating their environmental value proposition in relation to their impact.

Common anti-patterns:

  • Focusing solely on model accuracy without considering environmental trade-offs.

  • Implementing ML systems without measuring their carbon footprint.

  • Overprovisioning resources for ML workloads.

  • Unnecessarily retraining models when not required.

Benefits of establishing this best practice:

  • Improved alignment between ML initiatives and organizational sustainability goals.

  • Enhanced ability to meet regulatory and reporting requirements.

  • Transparent assessment of the sustainability impact of ML projects.

  • Identification of opportunities to optimize ML systems for lower environmental impact.

Level of risk exposed if this best practice is not established: High

Implementation guidance

Understanding the environmental impact of your ML workloads is essential for sustainable AI development. ML operations can be resource-intensive, requiring substantial computing power for training and inference, which translates to energy consumption and carbon emissions. By measuring and tracking this impact, you can make informed decisions about architecture choices, region selection, and optimization strategies.

The cloud provides significant advantages for sustainable ML development through its economies of scale and renewable energy investments. By leveraging AWS's efficiency, you can reduce your carbon footprint compared to on-premises alternatives. However, you still need to make conscious decisions about how you design, deploy, and operate your ML workloads.

Consider the full lifecycle of your ML system, from data collection and processing to model training, deployment, and ongoing operations. Each stage has different resource requirements and environmental impacts. For example, training large models is typically more computationally intensive than inference, but if your model serves millions of predictions daily, the inference stage might have a larger cumulative impact over time.

Implementation steps

  1. Establish sustainability objectives for ML workloads. Begin by defining specific sustainability goals for your ML projects that align with your organization's broader environmental commitments. Determine what metrics you'll track (for example, carbon emissions per training run or energy efficiency per inference) and set concrete targets for improvement.

  2. Assess the environmental impact of data processing. Evaluate how much data you're storing and processing for your ML workloads. Consider implementing data lifecycle management using Amazon S3 Lifecycle policies to automatically transition infrequently accessed data to more efficient storage classes or archive data that's no longer needed for active training.

  3. Measure the impact of model training. Calculate the computational resources and associated carbon emissions of your model training processes using tools like the AWS Customer Carbon Footprint Tool. Consider factors such as instance types, training duration, and region selection. Track these metrics over time to identify trends and opportunities for improvement.

  4. Optimize model training frequency and approach. Determine how often retraining is truly necessary based on model drift monitoring. Implement incremental training where possible using Amazon SageMaker AI to update existing models with new data rather than retraining from scratch. Use techniques like transfer learning to leverage pre-trained models and reduce computational requirements.

  5. Assess inference efficiency. Evaluate the environmental impact of your deployed models in production. Consider techniques like model compression, quantization, and distillation to reduce the computational requirements of inference while maintaining acceptable accuracy. Use Amazon SageMaker AI Neo to automatically optimize models for specific deployment targets.

  6. Calculate the net sustainability value. Compare the environmental impact of your ML workload with its benefits, including the sustainability improvements it enables. For example, if your ML system optimizes energy usage in manufacturing processes, calculate the net reduction in emissions to demonstrate its overall positive impact.

  7. Implement ongoing monitoring and reporting. Establish a regular cadence for reviewing sustainability metrics and reporting on progress toward your goals. Use AWS CloudWatch to monitor resource utilization and create dashboards to track your sustainability KPIs over time.

  8. For GenAI workloads, consider your model customization strategy. When implementing generative AI solutions, evaluate whether fine-tuning existing foundation models like Amazon Bedrock models is more environmentally efficient than training custom models from scratch. Consider prompt engineering and retrieval-augmented generation (RAG) approaches that can achieve business goals with lower computational intensity than full model training.

Resources

Related documents: