Project Overview
 |
MLEvol develops Software Engineering methods, practices, and tools to foster the continuous and efficient evolution of Machine Learning Systems (MLS), adapting their MLOps life cycle to a fast-moving ML ecosystem. |
Motivation
The current Machine Learning ecosystem grows at a remarkable pace, producing thousands of new assets each day: datasets, pre-trained models, pipelines, platforms, and supporting tools. However, the increasing heterogeneity and volatility of these resources make their integration into Machine Learning Systems (MLS) a complex, error-prone, and often costly activity. Maintaining quality, trustworthiness, and compliance while updating and evolving MLS in such a dynamic environment requires new Software Engineering approaches.
MLEvol addresses this challenge by treating the ML ecosystem itself as a driver for system evolution. It proposes a holistic, tool-supported framework that enables continuous adaptation, combining data-driven decision making, architectural mechanisms for automated change propagation, and governance strategies aligned with emerging AI regulations. The goal is to ensure that MLS can evolve sustainably and efficiently as their surrounding ecosystem evolves.
Objectives
Main objective: To develop Software Engineering (SE) methods, practices, and tools to foster the continuous and efficient evolution of MLS and the subsequent adaptation of their MLOps life cycle, considering the highly dynamic ML ecosystem.
Subobjectives:
O1 — ML Ecosystem Model & Asset Selection
Define, implement, and evaluate SE methods, practices, and tools to create a unified model for the highly dynamic ML ecosystem, enabling efficient selection of ML assets.
O2 — Data & Modeling Evolution
Define, implement, and evaluate SE methods, practices, and tools to enhance MLS evolution in the Data Management and ML Modeling stages within MLOps.
O3 — Integration & Operations
Define, implement, and evaluate SE methods, practices, and tools to efficiently integrate ML assets at the Software Development and System Operation stages within MLOps.
O4 — Governance & Compliance
Define, implement, and evaluate a flexible end-to-end holistic governance framework that facilitates the selection and integration of ML assets into evolvable MLS while considering MLS requirements and emerging AI-related regulations.
Share: