– theoretical discussions on how mathematical statistics or recent
mathematical models can be applied to learn lighter architectures
      – tutorials and experience feedback on how softwares like RAPL and
nvidia-smi can measure the energy consumed by deep learning algorithms.
      – A COMPETITION where participants optimize both their energy
consumption and the generalization ability of their models over the
CIFAR-10 dataset with common GPU architectures.

https://greenai-uppa.github.io/power_efficient_deep_learning/

Neural networks (NN) have become the most used family of machine
learning algorithms. Among the universality of architectures emerging
now, NNs are still prohibitive regarding environmental impact due to
electric consumption.
In this workshop, we expect to address these issues, based on both a
theoretical and practical deep learning analysis of standard pipeline
and new paradigms. The main theoretical discussions lie on how
mathematical statistics or recent mathematical models can be applied to
learn lighter architectures in order to reduce training and inference.
We also propose to use recent softwares like RAPL and nvidia-smi in a
dedicated tutorial and competition where the energy consumed by deep
learning algorithms will be measured and reduced.

CALL FOR PAPER:

We invite contributions to the 2021 workshop on power efficient deep
learning, and welcome paper submissions on artificial intelligence,
power consumption, AI carbon footprint and related areas. Candidates
must submit a two page proposal on the topic of their intervention. Top
contributions will be selected by a board of reviewers, and they will be
chosen to present their work at the workshop.

This workshop is a non-archival venue and there will be no published
proceedings. The papers will be posted on the workshop website. It will
be possible to submit to other conferences and journals both in parallel
to and after this workshop. At least one author from each accepted paper
must register for the workshop. Please see the ACML 2021 registration.

submission deadline: 05/11/2021

CALL FOR COMPETITION

Participants will optimize both their energy consumption and the
generalization ability of their models over the CIFAR-10 dataset with
common GPU architectures.

submission deadline: 08/11/2021