Frontiers in Machine Learning: Climate Impact of Machine Learning

Microsoft has made an ambitious commitment to remove its carbon footprint in response to the overwhelming urgency of addressing climate change. Meanwhile, recent advances in machine learning (ML) models, such as transformer-based NLP, have produced substantial gains in accuracy at the cost of exceptionally large compute resources and, correspondingly, carbon emissions from energy consumption. Understanding and mitigating the climate impact of ML has emerged at the frontier of ML research, spanning multiple areas including hardware design, computational efficiency, and incentives for carbon efficiency.

The goal of this session is to identify priority areas to drive research agendas that are best-suited to efforts in academia, in industry, or in collaboration. We aim to inspire research advances and action, within both academia and industry, to improve the sustainability of machine learning hardware, software and frameworks.

Session Lead: Philip Rosenfield, Microsoft

Speaker: Nicolo Fusi, Microsoft
Talk Title: Opening Remarks

Speaker: Emma Strubell, Carnegie Mellon University
Talk Title: Learning to Live with BERT

Speaker: Vivienne Sze, Massachusetts Institute of Technology
Talk Title: Reducing the Carbon Emissions of ML Computing – Challenges and Opportunities

Speaker: Diana Marculescu, University of Texas at Austin
Talk Title: When Climate Meets Machine Learning: The Case for Hardware-ML Model Co-design

Q&A panel with all 4 speakers

Date:
Haut-parleurs:
Philip Rosenfield, Nicolo Fusi, Emma Strubell, Vivienne Sze, Diana Marculescu
Affiliation:
Microsoft Research, Carnegie Mellon University, Massachusetts Institute of Technology, University of Texas at Austin