SustaiNLP 2020 Launches Competition to Create Effective Energy-Efficient Models for Difficult NLU Tasks
Energy efficiency has been a high priority in the scientific community for quite some time. The question remains: how do we perform the same tasks while utilizing the least amount of energy necessary? The SustaiNLP 2020 Shared Task Competition, led by NYU Computer Science PhD student Alex Wang and organized in part by CDS Professor Sam Bowman, hopes to help answer that question — at least in terms of AI and NLU (natural-language understanding). The organization has launched an industry-wide competition (shared task) to initiate more energy-efficient, language-based AI applications — i.e. contemporary AI and other technology tools (like Alexa for example) that would maintain the same effectiveness while only using a fraction of the energy. The competition is “centered around the SuperGLUE benchmark (a CDS-led project), which tests a system’s performance across a diverse set of eight NLU tasks. In addition to the standard SuperGLUE performance metrics, the competition will evaluate the energy consumption of each submission while processing the test data.” (SustaiNLP 2020 website)
Participants should submit their own trained models and model code, however the use of pretrained models and existing libraries is also allowed. SustaiNLP will use the experiment-impact-tracker library (Henderson et al. 2020) to measure the performance of each submission by evaluating energy efficiency via energy consumption. Competition submissions are due on August 28, 2020.
SustaiNLP’s workshop to promote both simpler and more sustainable NLP practices and research will take place on November 11, 2020 at EMNLP 2020 (the 2020 Conference on Empirical Methods in Natural Language Processing). The workshop was originally to take place in Punta Cana, Dominican Republic, but will now take place online due to the current COVID-19 pandemic. The workshop has two main objectives: the first is to encourage the development of more efficient NLP models and the second is to provide “simpler architectures and empirical justification of model complexity.” (SustaiNLP website).
For more information on the competition (shared task) and the EMNLP 2020 workshop, please visit the SustaiNLP 2020 webpage.
By Ashley C. McDonald
