A3: Machine Learning and Applications

Machine Learning is a scientific discipline concerned with the design and development of algorithms that improve their behaviour from experience (observations, labelled or not). It borrows techniques and theoretical background from the field of Artificial Intelligence, Logics, Statistics, to form a highly challenging domain. Machine Learning is now a mature field of computer science, with solid theoretical models and results, and a broad range of applications, both in industry and in multi-disciplinary research. It also became quite popular to a wide audience, thanks to Deep Learning successes in Game playing and with the emergence of Data Science, following the prominence of massive datasets requiring intelligent analysis.

Our group covers a remarkably broad range of topics, from Statistical Learning to Graph Mining and Reinforcement Learning. This allows us to be quite reactive to cope with new challenges raised by emerging applications of Machine Learning. It also makes it possible to study innovative combinations of learning methods for tackling complex problems.

The team A3 was created in 2005, with the goal of gathering Machine Learning related research of LIPN within a strongly coherent group, highly visible on the national and international research scenes. The structure of the group has evolved over the years, and now consists of four research axes, for which the group is widely recognized:

  • Machine Learning for Decision Support, which gathers research that deals with supervised learning related to decision support topics;
  • Collaborative Unsupervised Learning for Knowledge Transfer which studies non supervised learning in the context of distributed data, and transfer from one learning task to another.
  • Learning from Graphs, motivated by data mining issues, with a strong focus on mining structures attributed graphs and community detection and link prediction from complex (social) networks.
  • Learning topological models from massive datasets, which studies unsupervised learning of the hidden topology in the data, with a strong focus on online and massively distributed learning techniques.

In each of these axes, theoretical issues are addressed, and algorithms and software are developed, in the context of collaborative projects, including both academic and industrial partners. The team publishes in main Machine Learning international conferences (IJCNN, ICML, AAMAS to name a few) as well as main international journals (Artificial Intelligence, Machine Learning, Pattern Recognition, . . . ).

The strong links of the group with industrial partners is in particular testified by the high number of PhD theses in the group that take place in an industrial context, and its participation in FUI contracts. A member of the group has also created in 2010 a technological spin-off of University of Paris 13, that develops services in Intelligent Data Analysis such as Data Mining, Knowledge Discovery in Databases and Predictive Analytics.

The research topics of the team for the next five years aim at investigating promising areas where the team may emerge as a leader. These promising research topics are the following:

  • Unsupervised learning in the framework of optimal transport theory,
  • Deep Learning and quantum approaches
  • Mining augmented and heterogeneous graphs
  • New challenges for scalable computational statistics