Towards Architecting Sustainable MLOps: A Self-Adaptation Approach

TL;DR

The paper addresses sustainability challenges in MLS-driven systems, where data drift and evolving requirements lead to excessive energy use, costs, and reliability concerns. It proposes a self-adaptive MLOps architecture built on the MAPE-K loop and a design-time Decision Map to enable runtime responses to uncertainties across environmental, technical, social, and economic dimensions. The main contributions include integrating runtime adaptation into MLOps for sustainability, encoding design-time goals via a Decision Map, and validating the approach with a Smart City AQI forecasting case that demonstrates improved predictive performance ($R^2$ rising from $0.90$ to $0.94$) while reducing energy usage. These results suggest a practical path to more sustainable MLS deployments by combining monitoring, planning, and execution with structured knowledge and tactics. Future work will generalize the approach to other domains, broaden the uncertainty set, and explore applicability to unsupervised, reinforcement learning, and generative AI contexts.

Abstract

In today's dynamic technological landscape, sustainability has emerged as a pivotal concern, especially with respect to architecting Machine Learning enabled Systems (MLS). Many ML models fail in transitioning to production, primarily hindered by uncertainties due to data variations, evolving requirements, and model instabilities. Machine Learning Operations (MLOps) offers a promising solution by enhancing adaptability and technical sustainability in MLS. However, MLOps itself faces challenges related to environmental impact, technical maintenance, and economic concerns. Over the years, self-adaptation has emerged as a potential solution to handle uncertainties. This paper introduces a novel approach employing self-adaptive principles integrated into the MLOps architecture through a MAPE-K loop to bolster MLOps sustainability. By autonomously responding to uncertainties, including data, model dynamics, and environmental variations, our approach aims to address the sustainability concerns of a given MLOps pipeline identified by an architect at design time. Further, we implement the method for a Smart City use case to display the capabilities of our approach.

Figures (4)

• Figure 1: Flow diagram for our approach
• Figure 2: Approach
• Figure 3: Decision Map for AQI Prediction Pipeline
• Figure 4: $R^2$ Score & Log of Average CPU Consumption over $10s$ ($\mu J$). RT=retraining