ExperimentLens

Exploring, Monitoring, and Explaining complex AI pipelines

Live Demo

ExperimentLens dashboard: analyzing configuration effects on experiment performance.

About ExperimentLens

ExperimentLens is a lightweight yet powerful visual dashboard for the interactive exploration, monitoring, and explainability of complex AI pipelines.

Developed within the context of the ExtremeXP project, ExperimentLens empowers researchers, data scientists, and engineers to make sense of experimental results across multiple runs and interconnected tasks.

The tool is centered on human-in-the-loop experimentation, enabling users to monitor pipeline lifecycles, inspect results, and gain insights into pipeline behavior and configuration sensitivity.

Core Features

Run and Configuration Analysis

  • Compare metrics and outputs across multiple experiment runs.
  • Analyze the effect of variability points (e.g., hyperparameters, engine configurations).
  • Detect patterns, trends, and outliers across different pipeline variants.
Comparative analysis in ExperimentLens: visualizing metric distributions across pipeline variants.

Comparative analysis in ExperimentLens: visualizing metric distributions across pipeline variants.

Pipeline Control and Human-in-the-Loop Feedback

  • Monitor and steer pipeline executions at runtime, guided by intermediate results and visual feedback.
  • Use diagnostic views to adjust configuration choices and re-launch modified pipelines.
  • Support interactive refinement by integrating user feedback into the exploration process.
Model Performance Insights view in ExperimentLens

Model Performance Insights view in ExperimentLens

Artifact and Data Inspection

  • Preview input and output datasets, prediction results, and artifacts in rich visual formats.
  • Context-aware inspection scoped to the task and pipeline level.

Integrated Explainability

  • Access both local and global post-hoc explanation techniques.
  • Visualize results of explainability methods like counterfactual examples, Partial Dependence Plots (PDP), and Accumulated Local Effects (ALE).
  • Link pipeline behavior to specific inputs, parameter settings, and structure.

Technical Overview

Experiment tracking integration: Supports tools like MLflow for logging run-level metrics, parameters, and artifacts—enabling visual analysis of experiment outcomes.

Workflow orchestration integration: Connects with workflow orchestration frameworks to enable lifecycle control, task-level tracing, and dynamic adjustment of pipelines. Currently, the ExtremeXP Execution Engine is supported. Future integration is planned with additional tools such as Kubeflow, Airflow, and other orchestration platforms.

Get Involved

We welcome collaborations, feedback, and pilot use cases across domains.

To learn more, contribute, or request a demo:

Email

GitHub Repositories