Maximilien Danisch

PhD candidate at complexnetworks.fr,
Laboratoire d'informatique de Paris 6 and
Universite Pierre et Marie Curie.

Adress: 25-26-315, 4 place Jussieu, 75005, Paris.
Tel: 01 44 27 88 45 or 06 26 80 11 83
Email: maximilien.danisch@gmail.com

photoID

ABOUT ME

RESEARCH PUBLICATIONS

SPECIALS

ABOUT ME

I've been a Ph.D candidate in computer science at complexnetworks.fr under the supervision of Jean-Loup Guillaume and Benedicte Le Grand since September 2011, my focus topic is community detection in complex networks. In 2010, I obtained a master's degree in physics from ENS Cachan after an internship on granular matter at the City College of New York under the supervision of Hernan A. Makse. In 2010-2011 I spent one year working on machine learning problems as an intern at Columbia University under the supervision of Tony Jebara.

Curriculum vitae: pdf.

RESEARCH

  • Complex Networks

    I am investigating the community structure in complex networks. At a time where the vision of community structure is transiting from partition (simple but too unrealistic) towards overlapping communities (realistic but too difficult), I think that focusing on communities related to one node, i.e., ego-centered communities, is a good compromise. For this problem of ego-communities detection, I suggest to use a node-similarity approach rather than the only method investigated so far: a cost-function approach suffering from local minimums and hidden scale parameter. I also defined multi-ego-centered communities, i.e., communities related to a set of nodes: the key idea is that, although one node generally belongs to numerous communities, a small set of appropriate nodes, say two, can fully characterize a single community. A paper on this subject has recently been published in IJWBC 2012.

    Softwares:

    • A python code, CarOp.py, calculating the caryover opinion for a given node as explained in this paper.
    • The Louvain Method implemented in C by Jean-Loup Guillaume and Etienne Lefbvre as detailed in Arxiv.
    • Gephi, an open-source software for visualizing and analyzing large networks.
    • NetworkX, a Python language software package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks.

    Links:

    complexnetworks.fr - Hernan A. Makse - Eric Fleury - Luciano da F. Costa - Alain Barrat - Renaud Lambiotte - Cristopher Moore - Jure Lescovec - Emmanuel Viennet - Camille Roth - Barabasilab - Jon Kleinberg - Santo Fortunato - Marc Boullé - Nicolas Dugué - Gergely Palla - Illés Farkas -


  • Machine Learning

    Under the supervision of Prof. Tony Jebara, I worked on the applications of perfect graphs and semidefinite programing for Machine Learning. Many theoretical computer science problems are NP in general graphs, but become P when solved on perfect graphs using semidefinite programing. Some examples are: graph coloring, maximum clique, maximum stable set, vertex covering. These problems have direct applications for practical cases, e.g., scheduling, finding the maximum a posteriori... I thus think that two approaches should be considered to solve these theoretical problems in general graphs: (i) directly applying polynomial algorithms giving an exact solution on perfect graphs to general graphs in order to obtain an approximate solution possibly guaranteed, (ii) mapping a general graph into a perfect graph, then apply a polynomial algorithm to obtain an exact solution to the problem of interest on the perfect graph and thus hopefully obtain an approximate solution on the general graph, possibly guaranteed.

    Programs:

    • SDPLR, a C package and Matlab interface by Samuel Burer for solving large-scale semidefinite programming problems.

    Links:

    Tony Jebara - Yann Le Cun - Francis Bach - Andrew Ng - Samuel Burer - Laszlo Lovasz - Maria Chudnovsky -

  • Granular Matter

    In three dimensions, the maximum density of a packing of (same size) balls, 74%, is obtained for the face-centered cubic crystallized configuration. However, when shacking a packing of balls, the density of 74% cannot be obtained. The packing jams at a density in between approximately 63.5% and 53.5% depending on the friction between balls. I think that it is interesting to understand this property and to investigate what happens when the balls have different sizes or when they are deformed and also what happens in higher dimensions. This problem has applications ranking from the storage of grains to error correcting codes. When working under the supervision of Prof. Hernan A. Makse, I developed models to predict the phase diagram of packing of different size balls and rods, this work lead to two articles.

    Programs:

    • A python code, binball.py, calculating the volume fraction of binary packings of hard spheres following the framework explained in Arxiv.
    • Python codes voro2d.py, voro3d.py, calculating the voronoi boundary between two spherocylinders in 2d and 3d as detailed in Arxiv.
    • Fortran programs by Aleksandar Donev to generate and analyze two- and three-dimensional hard-particle packings using event-driven molecular dynamics.

    Links:

    Hernan Makse - Salvatore Torquato - Aleksandar Donev - Francesco Zamponi -

PUBLICATIONS

  1. Une approche à base de similarité pour la détection de communautés egocentrées. (FRENCH)
    M. Danisch, J.-L. Guillaume and B. Le Grand.
    ALGOTEL2013. PDF.

  2. Unfolding ego-centered community structures with "a similarity approach".
    M. Danisch, J.-L. Guillaume and B. Le Grand.
    CompleNet2013. PDF, PDFslides.

  3. Déplier les structures communautaires egocentrées - une approche à base de similarité. (FRENCH)
    M. Danisch, J.-L. Guillaume and B. Le Grand.
    FGG-EGC2013. PDF.

  4. Towards multi-ego-centered communities: a node similarity approach.
    M. Danisch, J.-L. Guillaume and B. Le Grand.
    Int. J. of Web Based Communities (2012). Team website.

  5. Jamming of hard rods I: From Onsager to Edwards.
    M. Danisch, A. Baule and H. A. Makse.
    (to appear). Arxiv.

  6. Model of random packings of different size balls.
    M. Danisch, Y. Jin and H. A. Makse.
    Phys. Rev. E. 81, 051303 (2010). Arxiv.

SPECIALS

  • My facebook scocialgraph
  • Free online classes, mostly computer-science/science, but not only.
    So far I followed:
    • Machine Learning, by Andrew Ng.
    • Introduction to Artificial Intelligence, by Sebastian Thrun and Peter Norvig.
    • Model Thinking, by Scott E. Page.
    • Game Theory, by Matthew O. Jackson and Yoav Shoham.
    • Algorithms: Design and Analysis, Part 1, by Tim Roughgarden.
    I enjoyed all of them and highly recommend them.