Postdoc position: information diffusion in graphs. INRIA, Lille.

The Mostrare team at INRIA-Lille (
invites applications for a full-time post doctoral position in the
area of machine learning for dynamic structured data, starting
(ideally) in June 2012.

Mission :

The research project will be focused on mining data represented as
graphs. Graph data arises in a wide variety of disciplines; for
example, social networks, heterogeneous databases and biosciences.
In this kind of graphs, aspects like multi-modality, dynamicity, or scalability raise new
important challenges that machine learning algorithm has to face.
Information diffusion on graphs consists on the probation of rumors,
news, labels from seed nodes to the rest of the graph.
Many works in the literature have considered
the problem of information diffusion in various domains like social
science, epidemiology, web, physics, marketing and new
applications ranging from advertising or recommendation, community
detection. Many of these works consider the problem of fitting a
diffusion model and try to explain the diffusion process.

In a first step of this work we will restrict ourselves to predict a
final step of this process from a given initial one or vice-versa
finding an initial state observing the result of the final diffusion
process. We will see that important tasks such as classification or
node labeling, link prediction in the setting of semi-supervised
learning can be casted as the information diffusion process. The
generalization will come by defining proper embeddings of the graph
where machine learning algorithms will learn the diffusion process for
the task under consideration.

In a second step, the candidate will also study efficient learning
algorithms and models that not only deal with large sizes of graphs,
but can adapt to the shifting trends of their dynamicity. Methods to
investigate include incremental learning, active learning or sampling
methods among others.

Finally experiments will be conducted on large graphs representing
brain activities and also on other heterogeous databases.

Skills and profile :

Applicants must have already or be very close to obtaining a PhD in
computer science, graph algorithms, machine learning. Some experience
in implementation and experimentation is expected. Fluency in English
is an important added-value. Informal enquiries may be directed by
email to: Gemma Garriga at or Marc Tommasi at