GENIAL

Tuesday

18.11.14

11:15 am

Vincent GRIPON

Telecom Bretagne

France

Error correcting graphs and long term memory

We start with three simplifying hypothesis on brain functionning: a) brain memory is redundant, b) brain can be seen as a recurrent neural network and c) storage of pieces of information in the brain is located at the scale of populations of neurons. Using principles of error correcting codes, we then propose a very simple model for information storage and retrieval thanks to cliques in neural graphs. We show how this model can be enriched to address complex problems and its connections to actual neuroanatomical knowledge.

Friday

28.06.13

3 pm

Nathanael Perraudin

LTS2 EPFL

Paper: Learning from Labeled and unlabeled data on directed graph, Zhou, Huang, Schölkopf

http://research.microsoft.com/en-us/um/people/denzho/papers/LLUD.pdf

Friday

05.07.13

2 pm

Li Pu

LIA EPFL

Clustering and Link Prediction with Hypergraphs.

Abstract: In many applications we are interested in clustering the entities by examining the associated relations, or predicting additional relations based on the existing ones. Hypergraphs can be used to represent the higher-order relations that are hard to describe with graphs, and serve as the model for clustering and link prediction problems. In this talk we discuss the clustering and link prediction problems in two scenarios. In the first scenario we present a new hypergraph Laplacian for clustering based on the minimum vertex-separator. And in the second scenario we examine the link prediction problem in recommender systems. 

Friday

12.07.13

2 pm

Mohamed Kafsi

LCA4 EPFL

The entropy of conditional Markov trajectories

http://arxiv.org/abs/1212.2831

Thursday

18.07.13

2 pm

Friday

26.07.13

2 pm

Pierre Vandergheynst

LTS2 EPFL

Monday

29.07.13

2 pm

Chloé Azencott

Univ. Tuebingen, Germany

Thursday

03.10.13

11 am

Yuxuan Li,

LUTS, EPFL

Congestion and propagation in urban transportation networks

It has been recently shown that a macroscopic fundamental diagram (MFD) linking network flow, density and speed exists in the urban transportation networks. This collective behavior concept can be utilized to introduce simple control strategies to improve mobility in homogeneous city centers without the need for details in individual links. However many real urban transportation networks are heterogeneous with different levels of congestion. In order to study the existence of MFD and the feasibility of simple control strategies to improve network performance, we firstly study the partitioning of transportation networks based on the spatial features of congestion for a static time period. Secondly, we explore empirical data from large-scale urban networks and reveal the hidden information during the process of dynamic congestion formation. Finally, we study some metrics on network structures that are strongly linked to the congestion propagation capabilities.

Thursday

17.10.13

2 pm

Yannick Rochat,

Digital Humanities,

EPFL and UNIL

Title: A social network analysis of Rousseau's Les Confessions

Abstract: We propose an analysis of the social network composed of the characters appearing in Jean-Jacques Rousseau's autobiographic Les Confessions with existence of edges based on co-occurrences. This work consists of twelve volumes, that span over fifty years of his life.

Tuesday 05.11.13

11 am

Michael Bronstein

USI, Lugano

Switzerland

Wednesday

13.11.13

at 11 am

Nicolas Tremblay

ENS Lyon

France

Wavelets on graphs: a tool for multiscale community mining in graphs

Abstract: For data represented by networks, the community structure of the underlying graph is of great interest. A classical clustering problem is to uncover the overall “best” partition of nodes in communities. We work on a more elaborate description in which community structures are identified at different scales. To this end, we take advantage of the local and scale-dependent information encoded in graph wavelets and classify nodes according to their wavelets or scaling functions. I will show results obtained on a graph benchmark having hierarchical structure and on a real social network.

Tuesday

19.11.13

11 am

Xavier Bresson

CHUV

Total Variation Data Analysis

Abstract: With enormous and daily flows of data in finance, security, health, social network and multimedia (sound/text/image/video), there is a strong need to process information as efficiently as possible for smart decisions to be made. Machine Learning develops analytical methods and strong algorithms to deal with this massive large-scale, multi-dimensional and multi-modal data. This field has recently seen tremendous advances with the emergence of new powerful techniques combining the key mathematical tools of sparsity, convex optimization and relaxation methods. In this talk, I will present how these concepts can be applied to find excellent approximate solutions of NP-hard balanced cut problems for unsupervised data clustering, significantly overcoming state-of-the-art spectral clustering methods including Shi-Malik's normalized cut. I will also show how to design fast algorithms for the proposed non-convex and non-differentiable optimization problems based on recent breakthroughs in total variation optimization problems borrowed from the compressed sensing field. This new total variation clustering technique paves the way to a new generation of learning algorithms that can provides simultaneously accurate, fast and robust solutions to other fundamental problems in data science such as Support Vector Machine data classification. These new methodologies have a wide range of applications including data retrieval (search engines), neuroimaging (diseases detection and analysis) and social network analysis (community detection).