PhD Thesis in Multi-Agent Systems and Optimisation

We offer a fully-funded 3 year PhD position entitled "Multi-agent
coordination using distributed optimization based on
belief-propagation and dual decomposition - application to dynamic
prosumption".




Description
â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?


The information and communications technology (ICT) infrastructure is
evolving and becoming increasingly dynamic and distributed.
Considering the context of sensor networks, the internet of things or
smart power grids, the expected impact on our society is huge. The
general problem of dynamic resource allocation (data, energy, etc.)
between producers and consumers (or prosumers for those that do both)
becomes critical for these infrastructures.


Multi-agent optimization techniques are a natural fit for such
systems:


1) the optimization problem is distributed by nature,
2) the system is highly dynamic with many events emanating from both
   the infrastructure and agent/human activities (and they may switch

   frequently from producer to consumer, and vice versa),
3) the scale of the system calls for strong privacy, which cannot be
   provided by a centralized system.


In the past years, multi-agent systems have been developed by using
message passing algorithms inspired by statistical machine learning
methods. Belief propagation is for instance designed for MAP (maximum
a posteriori) inference in probabilistic graphical models. Its
efficiency (in time and number of iterations) made it possible to
solve, in real time, problems like optimal energy allocation and
automatic service composition in sensor networks. Other message
passing algorithms using dual decomposition of the problem have
similar characteristics.


However, these approaches have some limitations â?"e.g. optimality and
convergence on cyclic networks. Despite recent advances, message
passing is still challenging and remains an open problem. Moreover,
since we consider problems with high dynamics and time-dependencies
(e.g. due to the flexible structure of the network, due to real-time
updates coming from sensors), extending such techniques seems
promising thanks to their runtime efficiency which can be exploited to
handle dynamics (though repeated solving processus) and
time-dependencies (by adding new time-related dimensions and
constraints).


This thesis subject aims at producing new methods and algorithms for
distributed optimization in dynamic networks with cycles.  It will
involve the following steps, non-exhaustively:


1. formalizing the problem of dynamic resource allocation and doing a
   theoretical analysis of the existing algorithms,
2. proposing new algorithms to handle the cyclic and dynamic nature of
   the networks,
3. implementing these algorithms in the form of generalization and
   extensions of existing algorithms,
4. evaluating empirically the algorithms by applying them to real data
   coming from the SEAS project, and deploying it into a smart-objects
   platform at EMSE.




General Information
â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?â"?


Duration: 3 years


Location: Saint-Etienne, France, with a possible collaboration with
IIIA-CSIC in Barcelona (Spain)


Supervisors: Gauthier Picard (MINES Saint-Etienne, Laboratoire Hubert
Curien UMR CNRS 5516, [http://www.emse.fr/~picard/]), Rémi Emonet
(Université Jean Monnet, Laboratoire Hubert Curien UMR CNRS 5516,
[http://home.heeere.com/])


Collaboration: Juan Antonio Rodriguez-Aguilar (IIIA-CSIC,
[http://www.iiia.csic.es/~jar/]), Jesus Cerquides (IIIA-CSIC)




Agenda
â.Oâ.Oâ.Oâ.Oâ.Oâ.O


Application deadline: May 10th


Interviews: May 15th to 20th


Final decision: first half of June


Starting date: September/October 2015




Skills related to the subject
â.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.O


The successful candidate will have and be willing to develop the
following skills:


â?¢ machine learning (graphical models, belief propagation, message
  passing algorithms),
â?¢ multi-agent systems, distributed optimization,
â?¢ proficiency in programming (Java/Python, C for Arduino) and
  optimization software (CPLEX, Gurobi),
â?¢ autonomy and English proficiency.




Location
â.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.O


Saint-Etienne is located about 2 hours from the mediterranean sea and
2 hours from Alps slopes. Lyon city is at 50 km. Saint Etienne has
about 180.000 inhabitants including more than 20.000 students.
Surrounded by hills where hiking and mountain biking are significant,
Saint Etienne is also member of the Unesco Creative Cities Network for
design.  [http://en.wikipedia.org/wiki/Saint-Ã?tienne]




Contact and application
â.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.Oâ.O


The application should include a motivation letter (short but
pertinent), a CV, degrees and grades, and relevant publications if any
(or Master thesis).  Candidates are also encouraged to provide
letter(s) of recommendation and contact information to reference
persons.


Applications must be sent before May 10th 2015, candidates are
encouraged to send application earlier.  Applications should be sent
to both of the following addresses: gauthier.picardâ?"atâ?"emse.fr AND
remi.emonetâ?"atâ?"univ-st-etienne.fr
____________________________________________