Greener computing in 鈥榖ig science鈥� is possible鈥� if we change our data processing approach
Big science projects 鈥� like those exploring the universe 鈥� generate huge data volumes with a heavy carbon footprint. A 蝌蚪窝视频 team is testing AI to compress this data, cutting storage needs and reducing energy use and emissions.
蝌蚪窝视频 researchers have been testing AI-driven compression approaches, training models to recognise data files and design algorithms that remove or modify less important elements, therefore reducing the amount of data needed. An example could be a compressed MP3 file with inaudible components of audio removed, at no loss to the listener.
One tool, 鈥楤aler鈥�, works with an autoencoder 鈥� a type of neural network trained to decrease the number of dimensions of input data, making it smaller.
Caterina Doglioni, Professor of Particle Physics, explains: 鈥淭here are multiple avenues to reduce the computing resources we use. One is reducing the amount of data to be stored through data compression.鈥�
The team are also measuring the energy usage of Baler and other approaches, to identify optimisations that could foster more energetically sustainable, data-driven scientific practices.
Rosie Schiffmann, an undergraduate student in the research team, adds: 鈥淲ith Baler and data compression as an example, we鈥檙e giving researchers a way to track their computational 鈥榤etabolism鈥� and make it more efficient. Green computing isn鈥檛 a futuristic vision; it鈥檚 actionable today if we rethink how we store and process data.鈥�
The work in this project received funding from the European Union鈥檚 Horizon 2020 research and innovation programme and European Research Council (ERC) under Grant Agreements n. and .
Meet the researcher
The project is led by Caterina Doglioni, Professor of Particle Physics, together with supervisors James Smith (Postdoctoral Research Associate) and Michael Sparks (Senior Research Software Engineer). Within the University of 蝌蚪窝视频 team are PhD student Pratik Jawahar, Jack Goodsall and Rosie Schiffmann from the Physics & Astronomy internship program, Bradley Booth from DeepMind鈥檚 AI Fundamentals Summer Internship program, and Sakshi Kumar, a Google Summer of Code student, working with collaborators in the US, Sweden and Ukraine.