Consortium

Inria is a French public research establishment dedicated to computational sciences, and is the primary contact point for the French Government on the computational sciences, including artificial intelligence. With 8 research centres and its 183 project teams, Inria has a workforce of 2 400 scientists with an annual budget of 231 million euros, 25% of which coming from its own resources. Under its founding decree as a public science and technology institution, Inria's missions are to produce outstanding research in the computing and mathematical fields of digital sciences and to ensure the impact of this research on the economy and society in particular. Research at Inria covers the entire spectrum of computer science, including challenges raised by other sciences as well as socio-economic sectors. Inria's identity and strength are forged by its ability to develop a culture of scientific innovation, to stimulate creativity in digital research. In particular, Inria has been charged by the French Ministry of Research and Education to oversee the creation of a network of AI research institutes in France. Many Inria project teams participate in European framework programmes with 49 Horizon 2020 projects currently underway. INRIA is also one of the most successful research institutions in the ERC program, with 31 ERC projects currently underway within the 8 research centres. Inria research teams publish more than 5000 Scientific Journal and Conference Papers per year, with preference for publications in leading, high impact scientific Venues. Inria has an internal policy to promote open science, and requires its research teams to publish pre-print copies of all scientific publications in the HAL open archive system.

Inria will lead WP 4 on Interaction and User Studies.

The work will be led by Janin Koch, a recognized researcher specializing in Human-Computer Collaboration supported by Eya Ben Chaaben (PhD) working on human-computer exploration for sustainability. They will take advantage of their extensive experience in designing interactive systems for explorative processes to allow users to interactively explore the trade-offs of competing ML models together with intelligent agents, but always with the goal of providing control to the user.

Exploring ML model alternatives during the development process, before the models enter the training cycles, requires users to express potentially ambiguous project objectives and to understand the trade-offs of ML model alternatives, e.g. time, computing hardware, or estimated CO2 footprint for a particular task.

Based on interviews with ML experts we identify relevant factors of model selection in practice and develop new explainability approaches to allow intelligent systems to communicate alternative ML models within such a context-dependent process, while taking resource cost estimates into account.

This exploratory interface will allow ML developers better express, define, and develop relevant project descriptions for the SustainML pipeline and allow users to compare existing model trade-offs.