[27/09] Round Table
Artificial intelligence: varied and complementary skills
In the ten years since its creation, the Inria Bordeaux - Sud-Ouest research centre has seen its teams develop a large number of research projects in the extended sphere of artificial intelligence – to the point of creating an incredibly rich ecosystem. Guided tour.
Like Inria’s seven other research centres, the Inria Bordeaux - Sud-Ouest Centre is very active in the world of artificial intelligence, “but one of the specific features of our centre is to bring together a wide range of skills in all the subfields that make up the AI nexus or surround it, such as learning techniques, statistics, optimisation, high-performance computing and robotics, as well as neuroscience and cognitive sciences”, sums up Frédéric Alexandre, leader of project team Mnemosyne. And, in fact, of the centre’s 21 teams, more than half conduct research that is related to artificial intelligence. Broadly speaking, Inria Bordeaux - Sud-Ouest’s project teams can be divided into three main families. “The most significant of them brings together entities that work on learning techniques with a mainly mathematical basis, particularly from the world of statistics and probabilities,” explains Pierre-Yves Oudeyer, who leads the project team Flowers. Among them are CQFD, who are developing statistical and probabilistic methods for modelling and optimising complex systems, Sistm, a specialist in biostatistics, Geostat, who analyse multiple-scale complex signals, PLEIADE, who focus on digital biology, and project teams Carmen and Monc.
From high-performance computing…
Another area of activity historically explored at Inria Bordeaux is the field of high-performance computing (HPC), which is quickly becoming one of the essential building blocks for the development of artificial intelligence at a time when learning algorithms need to be able to manage petabytes of data from multiple sources,
says Frédéric Alexandre. In fact, most teams involved in HPC are now taking a greater interest in artificial intelligence. This is the case for Tadaam, who are conducting research in particular on the modelling of high-performance computing systems and applications, with the aim of understanding the interaction between machines and the applications that they run. It is also the case for HiePACS, a specialist in numerical simulation bordering on applications in the field of materials chemistry, for Cardamom, who are seeking to establish models to describe and anticipate fluid flows, and for Memphis, who are developing numerical models on high-performance computing infrastructure in order to meet industrial needs.
Each in its own way, Flowers and Mnemosyne have positioned themselves in a niche close to the vision of the pioneers of artificial intelligence, who wanted to give machines some of the functions of living beings.
“Within the Mnemosyne project team, we focus primarily on the different kinds of memory and on the mechanisms that enable our neural networks to combine these different kinds of memory and, in this way, to be efficient at learning,” continues Frédéric Alexandre.
The challenge is to contribute to medical research by fostering a better understanding of certain pathologies but also to contribute to improving the learning capacities of machines which are still a very long way away from our own!
“For their part, the researchers in Flowers want to understand how humans, in particular children, learn by focusing on the drivers of intrinsic motivation, beginning with curiosity,” says Pierre-Yves Oudeyer.
We are trying to model these human mechanisms and draw assumptions from them to enable machines to gain autonomy and a greater ability to interact with the real world.
This research has applications in a wide variety of fields: education, video gaming, assistive robots and ‘curious’ robots for the discovery of new molecules, amongst others.In a different field but still close to neuroscience, project team Potioc researchers are exploring new ways for the brain-computer interface to foster rich and fluid interactions with the digital world with projects ranging from the creation of an ‘augmented sandbox’ dedicated to meditation to a project to represent brain activity on a robotic puppet.
As we can see, the Bordeaux teams’ playgrounds are particularly diverse – but are they too disparate to make common cause? “On the contrary, we believe that this diversity is a strength,” says Frédéric Alexandre. “One of the ambitions of the Inria Bordeaux - Sud-Ouest Centre over the coming years is to create synergies – particularly on the subject of learning – to make the most of all of our expertise and to serve the general interest, particularly in the field of medicine where expectations as regards artificial intelligence in all its forms are very high.”
Thursday 27 September: celebrating 10 years of research and technology transfer
First highlight of the day on 27th September 2018: three successive round tables organised by Inria and hosted by ENSEIRB-MATMECA, a Bordeaux INP engineering school, from 4pm to 7.30pm. They will provide an opportunity to discuss major digital issues that are also important societal matters with researchers and industry professionals from around the world, who form the Inria Bordeaux - Sud-Ouest ecosystem.
Round table 2
Artificial intelligence, when machines learn
- Véronique Aubergé, researcher at CNRS
- Raphaël Charrier, founder and CEO of Qucit
- Jean-Gabriel Ganascia, Sorbonne University Professor
- Pierre-Yves Oudeyer, research director at Inria and leader of the Flowers project team