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.
Quick links
Energy Consumption Oriented ML Task Modeling
Knowledge binning catalog and efficient practise registry
by DFKI