Towards a new paradigm in artificial intelligence
Artificial intelligence, and in particular the field of statistical learning (machine learning), has made considerable progress in recent years. Image recognition, speech recognition and machine translation algorithms are examples of success stories that are already being implemented in everyday life. But AI’s current model has its limitations. It works in a very centralised way: each user provides his or her data, and then the calculations are done in one place. It poses technical (bottlenecks in data processing) and social (private data, power’s centralisation) problems.
OCEAN (On IntelligenCE And Networks) project aims to develop a new learning methodology where learning is shared between different users, each retaining their autonomy and control over their data. Funded by the European Research Council (ERC), to the tune of €10 million over 6 years, this project is a collaboration between interdisciplinary research teams from several institutions, led respectively by Éric Moulines, École Polytechnique, Michael Jordan, UC Berkeley, Christian Robert, Université Paris Dauphine-PSL and Gareth Roberts of the University of Warwick.
The starting point for this collaboration is federated learning, where several agents participate in the learning task while maintaining control over their data. This principle has been in full swing in recent years, for example, in "next word search" algorithms which suggest words to complete a phrase typed into a search bar (in this case, the learning is done without the queries being recorded and stored for this purpose). But we remain with a centralized paradigm to train models.
OCEAN intends to go further by focusing on some shortcomings of federated learning, such as dealing with uncertainty about the result given by an algorithm or the fact that some users may want to take advantage of other people's data without providing viable data themselves (free-riders), or that other users may have an interest in biasing the algorithm in their favour. Economic mechanisms must then be found to reward those who provide good quality data.
"Considering each user as an active agent with its own strategy is a new paradigm for artificial intelligence," explains Éric Moulines, Professor at École Polytechnique's Centre for Applied Mathematics (CMAP). In addition to learning, the aim is to study how to make decisions in a shared environment. Possible applications would be, for example, to agree on a market linking energy producers and consumers.
All of these questions open up avenues of research that are still very fundamental and remain to be explored. It is also an emerging topic where academic research can contribute important and original ideas in a very competitive field. OCEAN brings together leading specialists in mathematics and statistical learning and interdisciplinary teams (multi-agent systems, microeconomics and game theory, decision theory, etc.).
About the ERC synergy Grants:
Synergy Grants support small groups of two to four Principal Investigators to jointly address ambitious research problems that could not be addressed by the individual principal investigators and their teams working alone. The projects should enable substantial advances at the frontiers of knowledge, stemming, for example, from the cross-fertilisation of scientific fields, from new productive lines of enquiry, or new methods and techniques, including unconventional approaches and investigations at the interface between established disciplines. The transformative research funded by Synergy Grants should have the potential of becoming a benchmark on a global scale.
*CMAP : Applied Mathematics Center - joint research unit CNRS, École Polytechnique – Institut Polytechnique de Paris