Work Packages

SustainML Work Packages overview

The SustainML project is founded on the principle that reducing the CO2 footprint of machine learning (ML) applications requires making power-aware applications as straightforward to develop as standard ML systems are today. This initiative aims to empower users with limited knowledge of architecture choices and energy footprints to easily minimize the power consumption of their applications. We envision a sustainable, interactive ML framework for Green AI that will comprehensively prioritize and advocate for energy efficiency throughout the entire lifecycle of an application, thereby avoiding AI waste. The project has several key objectives: modeling the requirements of specific ML applications, developing resource-aware optimization methods based on these models, creating a transparent design assistant to guide developers, compiling a collection of energy-efficient methods and parameterized ML models, and implementing a dedicated toolchain to support these efforts.

WP1

Energy Consumption Oriented ML Task Modeling

Knowledge binning catalog and efficient practise registry

by DFKI

WP3

Energy Consumption Optimized ML Toolkit and Methods

Resource utilization aware NAS framework

by KU

Knowledge recycling scheme and fuctional knowledge core catalog

by IBM

WP5

Framework Architecture and Im- plementation

WP2

Hardware Architectures for Low Power ML

Open-source hardware exploration toolchain

WP4

Interaction and User Studies

Explainable AI for decision making

Working prototype of a ML-exploration tool and valuation

WP6

Application Oriented Validation

Use cases validations

by IBM

Logo ALMA

SustainML is among these nine innovative projects dedicated to creating a sustainable ML framework for Green AI.

Coordinator Office Address

Plaza de la Encina 10-11, Núcleo 4, 2ª Pl.
28760 Tres cantos - Madrid (España)

  • X
  • LinkedIn

EN-Funded_by_the_EU-POSThis project has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No 101070408.