SustainML is committed to reducing the carbon footprint in the ML processes. With this goal in mind, the partner project University of Copenhagen, has developed a framework for self-supervised image segmentation.
In the era of advancing technology, the need to address environmental concerns has become increasingly important. SustainML aims to contribute to reducing carbon dioxide (CO2) emissions and promote the research of green artificial intelligence (AI). As part of this initiative, Raghavendra Selvan, from the project partner University of Copenhagen, has developed a simple and efficient framework for self-supervised image segmentation using contrastive learning on image patches, developed in their paper Efficient Self-Supervision using Patch-based Contrastive Learning for Histopathology Image Segmentation.
The model consists of a fully convolutional neural network (FCNN) that predicts confidence maps indicating the likelihood of each pixel belonging to a specific object class. The FCNN is trained contrastively using positive and negative patches sampled from the images based on an entropy-based distribution that captures patch-level confidence scores. Convergence is assumed when the information separation between the positive patches is small and the positive-negative pairs are large.
The proposed model achieves the following:
- Efficient and straightforward framework for self-supervised image segmentation,
- Smaller algorithm (FCNN) model with faster convergence compared to relevant methods
- Comparable performance to supervised and self-supervised baselines.
The model's smaller size and faster convergence make it a valuable tool for researchers and practitioners aiming to reduce energy consumption and promote green AI. As part of the SustainML project, this framework contributes to developing sustainable AI practices and helps mitigate the environmental impact of AI technologies.
By Raghavendra Selvan from Machine Learning Section, Kiehn Lab & Data Science Lab, University of Copenhagen.
All the SustainML participants: DFKI, Inria, IBM, University of Copenhagen, UpMem, Technische Universität Kaiserslautern and eProsima.
MORE INFORMATION ABOUT SUSTAINML:
For any questions, please get in touch with
This project has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No 101070408.