New iCoSys research on energy-efficient Document AI

How much additional energy is required to improve the accuracy of Document AI systems – and at what point does the extra computational effort stop being worthwhile?

At ACM DocEng 2026, iCoSys researchers Adam Grossenbacher, Loïc Guibert, Beat Wolf and Jean Hennebert presented their paper, “More Power, More Accuracy? An Energy-Efficiency Analysis of Intelligent Document Processing.”

The research examines the trade-off between energy consumption and accuracy in Intelligent Document Processing (IDP), also known as Document AI. The topic is increasingly relevant as more computationally demanding AI models are used for document processing tasks.

Beyond the environmental impact of AI, energy efficiency can also have practical implications for operational costs and for applications deployed on battery-powered or resource-constrained devices.

About the paper

The study investigates how increasing computational effort affects both model accuracy and energy consumption, helping to better understand when additional computation delivers meaningful improvements and when the gains begin to diminish.

👏 Congratulations to Adam, Loïc, Beat and Jean on the publication!

🔗 Read the paper: https://lnkd.in/eWWsdkxf