Latest News

eProsima is proud to announce the release of SustainML v0.2.0, a major update to its backend and developer framework for sustainable machine learning. This new version strengthens stability, usability, and extensibility across both the backend and the Qt-based front-end, introducing smarter task management, improved data handling, and deeper integration with research partners and frameworks.

Highlights of v0.2.0

🚀 Enhanced Experiment Workflow and Results Management

This release delivers a smoother and more robust user experience when defining, iterating, and comparing ML problems:

• 10 Base Outputs: All tasks now default to 10 outputs for improved consistency and downstream handling.
• Screen Management Refactor: Screens are now persistent—no longer re-created each time—resulting in faster, more stable transitions.
• Preserved Inputs and New Problem Handling:
  • The Go Back button in Results now returns to Problem Definition while preserving all prior inputs.
  • A new “New Problem” button resets Problem Definition to a clean state.
• Improved Navigation: Open tabs remain visible when switching between screens.
• Iterative and Comparative Stability: Tabs now automatically close or refresh after reiteration to prevent data mismatches.
• Accurate Problem Tracking: Fixed reading of current_problem_id to avoid displaying wrong results when no active problem is selected.

📊 Smarter Visualization and Comparison Tools

• Dynamic Tab Names: Comparison tab names now reflect the specific variable being compared.
• Multiple Comparison Tabs: Support for several concurrent comparisons in a single session.
• Improved Graphs: Removed redundant table titles; units are now shown directly on the y-axis.
• Enhanced Iteration Details: “More Info” is now displayed in an interactive table instead of JSON.
• New Popups & Tooltips: Added More Info popups for result variables and tooltips for action buttons for a clearer user experience.
• Renamable Tabs: Problem tab names can now be customized directly in the interface.

🧠 Backend and Metadata Improvements

• Dataset Metadata Extraction: Automatic retrieval of dataset metadata from a given dataset path.
• Overview Result Reception: Improved handling of comparison data and waiting states, resulting in smoother backend responses and better stability.
• Carbon Footprint Tracking Restored: The CO₂ footprint is once again available in result screens.

🔌 Framework & Integration Enhancements

• RPTU Hardware Predictor Integration: The framework now integrates with RPTU hardware, providing latency measurements for predictions. Power consumption reporting will follow in future updates.
• DFKI RAG Integration: Enables advanced knowledge retrieval and reasoning through the DFKI RAG system.
• Neo4j Graph Backend: Query mechanisms have been upgraded from TTL to Neo4j, improving scalability and query expressiveness.
• Hugging Face & Submodule Loading: Models can now be downloaded directly from Hugging Face, with support for locally loading unsupported modules from files.
• New Executable: Added sustainml-framework as a standalone entry point for developers.

Summary

With v0.2.0, eProsima continues to expand the SustainML ecosystem—introducing smarter front-end state handling, deeper dataset and metadata integration, and richer backend interoperability with hardware and knowledge graph systems. This release focuses on making sustainable ML experimentation more interactive, stable, and reproducible, empowering researchers and developers alike.

For further information and to get started, visit the GitHub repositories
👉 SustainML_Framework and SustainML_Library
or consult the official SustainML documentation.

Written by Denisa Alexandru at eProsima.