IT and Telecom

Exploratory actions: targeting new research pathways

Date:
Changed on 20/01/2023
Inria’s exploratory actions programme was set up to promote the emergence of new research themes, giving scientists the means with which to test original ideas with a view towards forming Inria project teams.

Exploratory actions give us an opportunity to put our trust in the intuition of the members of our research teams. The aim of this scheme is to focus resources around highly innovative subjects, targeting approaches that are “off the beaten track”, risky and/or which represent a disruption in relation to traditional approaches.

The programme will make it possible to delve deeper into certain subjects and to prove their relevance from a scientific perspective, a vital stage in the build-up to forming a project team. This may also involve exploring themes that are unusual or marginal for Inria, such as subjects in the social sciences or in the legal domain. 

Providing a greater incentive for risk-taking in science

What makes a proposal “exploratory” is not easy to sum up in a few words, given that we could reasonably class all research as exploratory. Irrespective of the research domain, however, certain subjects or approaches are clearly better established, either through their lifespan or the number of researchers. Subjects may be applied or theoretical, rooted in digital or in interdisciplinarity, provided they remain faithful to the underlying principle for all Inria research: to focus on the impact the results could have in the long-term, either in computational science or in the fields in which they are used.

In 2023, 6 exploratory actions from the Inria center at the University of Lille were launched within the context of the programme.

AlaMVic, for optimised virtual machines

Stéphane Ducasse
Stéphane Ducasse - Inria / Photo B. Fourrier

Virtual machines (VMs) are ubiquitous in all laptops, servers and phones. Industrial VMs (e.g. from Microsoft, Oracle, Google...) use very elaborate optimisation techniques, often hand-crafted by experts, which are difficult to replicate, reproduce and modify. These optimisation techniques are mainly aimed at improving speed, and are incompatible with constraints such as space and energy efficiency, which are important in IoT or robotics.

In AlaMVic, we propose to address the construction of VMs using a holistic generative approach, in contrast to existing approaches that focus on speed and unique VM components such as the JIT compiler. We explore how to turn hand-crafted optimisations into generative heuristics, how they are applied and combined in domains such as IoT and robotics, and new methods and metrics for evaluating VMs in these domains.

Comanche, to learn and represent the meaning of words

COMANCHE proposes to transfer and adapt recent representation learning algorithms from deep learning in order to model the evolution of word meaning, by confronting them with theories on language acquisition and language diachrony. At the crossroads of machine learning, psycholinguistics and historical linguistics, this project will validate or revise some of these theories, but also allow the emergence of computational models that are more data and computationally efficient because they exploit new inductive biases inspired by these disciplines.

ETHICAM, the Internet of Everything to communicate

Valéria Loscri
Valéria Loscri - Inria / Photo C. Morel

After the Internet of Things (IoT), the Internet of Everything. This is the subject of the exploratory action led by Valeria Loscri. More precisely: how to overcome the limits of current means of communication in order to have connected and autonomous objects capable of talking to each other efficiently. The potential of radio frequencies is now being exploited to its fullest extent... but the needs in terms of throughput, speed and communication spectrum continue to grow.
The researcher therefore wanted to explore new avenues, to define new paradigms. The idea is to enable information exchange from nanodevices, objects, materials and even biological systems. For example, by looking at phonons, the equivalent in sound of the photons of light. A better understanding of the characteristics of matter and such particles could one day enable them to be used to transmit information more efficiently than at present.
However, this molecular communication is still a vast unknown. It is precisely for this reason that Valeria Loscri's studies fit perfectly into the exploratory actions, since they involve innovative, fundamental and risky research. The project, which is structured over three years, should first of all make it possible to create collaborations with other researchers in materials or physics in order to lay the foundations for interdisciplinary research. Secondly, it should enable the definition of new communication paradigms. Finally, to be able to imagine innovations, such as intelligent materials, capable of communicating with each other and which could, for example, be integrated into future autonomous cars.

FLAMED, artificial intelligence for health

Portrait A bellet
Aurélien Bellet - Inria/ Photo C.Morel

With the creation of FLAMED, Aurélien Bellet, a researcher in the Magnet research team, proposes to explore a decentralised approach to artificial intelligence applied to healthcare. In close collaboration with the INCLUDE team of the Lille University Hospital, the objective of Flames is to carry out data analysis and machine learning tasks involving several hospitals while allowing each site to keep its data in-house and guaranteeing their confidentiality. The research and software developments we intend to carry out aim to obtain a set of technical solutions for the construction of health systems that are respectful, by construction, of privacy.

 

 

PATH, for an adapted patient pathway

Sophie Dabo
Sophie Dabo - Inria

European healthcare systems are facing multiple challenges, including an ageing population, an increase in chronic diseases and patients suffering from multi-morbidity, and limited financial and human resources. The response to these challenges relies in particular on the organisation of care into care pathways, justified by the scientific literature and supported in France by political guidelines. The analysis of care pathways and their adequacy to needs and resources has thus become a major scientific and administrative challenge. Although the numerical data available for this purpose is increasing rapidly, the statistical methods and tools available to researchers and health authorities remain limited and inefficient. PATH proposes to develop statistical methods for the construction/analysis of the patient pathway through two applications dealing with the re-hospitalisation of the elderly and post-operative complications.

SR4SG gives a societal purpose to sequential learning

Odalric-Ambrym Maillard
Odalric-Ambrym Maillard - Inria

At present, sequential learning is mainly used to display targeted advertising on the Internet. This is far from satisfactory, according to Odalric-Ambrym Maillard. The researcher has therefore proposed an exploratory action that aims to give a societal dimension to this sub-field of machine learning.

The aim of this innovation?
To combine sequential learning, sustainable agriculture, soil conservation and biodiversity. In concrete terms, it involves creating a platform based on recommendation algorithms for the emergence of collaborative sharing of good agricultural practices. Today, various players, such as the Natural History Museum, CIRAD (Centre de coopération internationale en recherche agronomique pour le développement) and INRA (Institut national de la recherche agronomique) collect observation data on gardens. But they do not have the necessary machine learning skills to make personalised recommendations.
The first aim of this exploratory action is therefore to bring together specialists in biodiversity and good agricultural practices on the one hand, and Inria researchers in artificial intelligence on the other, in order to encourage the emergence of coherent research issues.
Secondly, work on data collection. Indeed, it is not simply a matter of obtaining observations but also a history of actions and effects, which is essential to enable sequential learning by the algorithms. This stage will lead to the creation of an application allowing gardeners, experienced or amateur, to share their observations, their actions and the effects observed.
The final objective of the exploratory action will then be to develop the algorithms that will use this data to learn and make recommendations adapted to each plant, in each garden, according to each context.
Of course, given the scale of the project, it is not envisaged that a perfect application will be deployed on a global scale in four years. However, Odalric-Ambrym Maillard does hope to have a prototype platform that brings together a mixed scientific community, which will then make it possible to obtain ANR-type funding to continue the experimentation. The researcher hopes that this exploratory action will lay the foundations for tomorrow's tools for good agricultural practices.