To main content

ExperimentLens: Interactive Visual Analytics and Explainability for ML Experiment Management

Abstract

The widespread adoption of experiment tracking and MLOps platforms has streamlined the management of machine learning workflows. Yet, these platforms often fall short in supporting interactive visual analysis that combines experiment results, data exploration, and model explainability within a unified interface. To address this gap, we introduce ExperimentLens, an extensible experiment analytics tool that operates on top of existing tracking infrastructures and supports multiple platforms through a simple adapter interface. ExperimentLens offers a rich, web-based environment for comparing runs, visualizing performance metrics, exploring datasets, and interpreting model outputs. Its modular architecture augments standard tracking systems with flexible, interactive capabilities that support both routine monitoring and in-depth analysis. We illustrate ExperimentLens’ functionality through a walkthrough of its architecture and user interface.

Category

Academic article

Language

English

Author(s)

  • Stavros Maroulis, Vassilis Stamatopoulos, Panagiotis Gidarakos, Konstantinos Tsopelas, Nikolas Masouras, Konstantinos Kozanis, Nikolas Theologitis, George Papastefanatos, Giorgos Giannopoulos, Erik Nilsson
  • Erik Gøsta Nilsson

Affiliation

  • SINTEF Digital / Sustainable Communication Technologies

Date

05.09.2025

Year

2025

Published in

Proceedings of the VLDB Endowment

View this publication at Norwegian Research Information Repository