Adopting innovative initiatives that seek a positive environmental impact is a challenge. To better understand this trend, we spoke with Lucía Bouza, collaborator and researcher at Pyxis, who has been involved in projects related to sustainability in technology for 3 years, and who in 2023 published How to estimate carbon footprint when training deep learning models? A guide and review.

Nowadays, sustainability and environmental care are relevant topics in several industries. Technology is no exception. In this context, «Green Software» has emerged as a response to the growing need to develop more sustainable and environmentally friendly technological solutions.

Green Software is the design, development and use of software in a way that minimizes its environmental impact. This encompasses various practices and strategies, such as optimizing the use of computational resources, improving energy efficiency and reducing the carbon footprint associated with the software life cycle.

• How did you become interested in investigating this practice?

My first approach to the Green Software concept was in Paris, while working under the direction of Aurélie Bugeau, a person who took these issues very seriously, from her daily life to her professional career. It was she who introduced me to the concept of Green Software, and encouraged me to do more research on the subject.

• What is the topic of the paper?

The research is entitled «How to Estimate Carbon Footprint When Training Deep Learning Models? A Guide and Review». This research was about the evaluation of different tools for carbon footprint estimation during the training of deep learning models. In the article, an introduction to the topic and a comparison of the tools is made. From the experiments, some tips for choosing the right tool and infrastructure are provided.

The paper also involved Cambridge researcher Loïc Lannelongue, who developed one of the tools studied .

• How have you applied your knowledge of Green Software at Pyxis?

Currently in the Cognitive Solutions unit of Pyxis we are migrating our services in datacenters located in Sweden. This decision was made to reduce the carbon footprint of our activity. We took into account the carbon intensity of the different regions where Microsoft datacenters are located in different time periods, the sustainability reports, as well as the availability of the different services and their cost. In addition, we also began to explore the reports available on the platform in order to measure our impact. 

Challenges and Opportunities

Uruguay has one of the cleanest energy sources in the world, thanks to an energy matrix that consists of a very high percentage of renewable energies. The opportunities we have as a country are enormous. However, we still face several challenges, such as the need to increase awareness of sustainability in the technology sector.

• What challenges can you face in the implementation of Green Software?

I think one of the biggest challenges is at the level of training, since the implementation of sustainable technologies is not in the curriculum of tertiary technology careers. Without information, it is very difficult to understand the concepts and make the right decisions.

Another challenge is the lack of awareness about the impact of our actions. It is important to understand how we generate impact, which we will always generate, and how to reduce it.

And finally, although Uruguay has a very good energy matrix, the use of cloud technologies is a fact, and unfortunately we do not have in Uruguay any datacenter of the main commercial clouds. So it is not so easy to take advantage of the virtues of our country in terms of energy for technological services, and we must carry out certain studies to choose datacenters abroad.

• What is your vision for the future of Green Software?

I believe that with small actions we can obtain significant improvements in the impact of our activity. With more education and information, we can raise awareness and make Green Software the norm rather than the exception.