Graph-structured datasets range from online social networks (OSNs), such as Facebook, LinkedIn, and Twitter, to biological networks, such as proteins-protein interaction (PPI) and gene regulatory networks. Data analytics techniques have the potential to extract knowledge and make predictions from such graph data. We develop stochastic models for large graphs, algorithms for their analysis, and fundamental results about their properties.
PROPER: global protein interaction network alignment through percolation matching H. Hassani, M. Grossglauser and H. P. Modarres BMC Bioinformatics, 2016.
The mobility traces generated by millions of users on the move with their smartphones is central in many sectors of the digital economy. Services built on mobility fuel a host of new business models, including location-based advertisement, navigation and transportation, and augmented reality. We explore machine learning problems and applications that exploit the rich structure in mobility data within a large population of mobile agents.
Uncovering Latent Behaviors in Ant Colonies M. Kafsi, R. Braunschweig, D. Mersch, M. Grossglauser, L. Keller and P. Thiran 2016 SIAM International Conference on Data Mining, Miami, Florida, USA, May 5-7, 2016.
Traveling Salesman in Reverse: Conditional Markov Entropy for Trajectory Segmentation M. Kafsi, M. Grossglauser and P. Thiran 2015 IEEE International Conference on Data Mining, Atlantic City, NJ, USA, November 14-17, 2015.
Hierarchical Routing Over Dynamic Wireless Networks D. Tschopp, S. Diggavi and M. Grossglauser Random Structures & Algorithms, 2015.
Epidemic models capture how infectuous processes propagate on graphs, such as a disease through a contact graph, or an idea through a social network. These models allow us to study the dynamics and long-term asymptotics of epidemic processes, and to explore measures to manage them, such as vaccination to slow the process, or deliberate infections to optimize the spread of an opinion. We are also particularly interested in estimation problems on epidemics under monitoring budget constraints.
A General Framework for Sensor Placement in Source Localization B. Spinelli, E. Celis and P. Thiran IEEE Transactions on Network Science and Engineering, 2017.
The effect of transmission variance on observer placement for source-localization B. Spinelli, E. Celis and P. Thiran Applied Network Science, 2017.
Back to the Source: an Online Approach for Sensor Placement and Source Localization B. Spinelli, L. E. Celis and P. Thiran 26th International World Wide Web Conference (WWW), 2017.
Large scales are inherent to modern information processing systems, which can no longer be designed to work in a centralized and controlled manner, but will need to cope with a strong random component and rely on principles of self-organization. We develop distributed algorithms for information processing, message passing, decision making, learning and optimization in large-scale networks whose nodes are often randomly deployed, have very limited functionalities, and may fail for various reasons.
Discrete choice models capture how we select one option among a set of alternatives. In the digital world, we are forced to make choices all the time: we click on a search result in a list, choose a song from iTunes, select a route to a destination, or pick a restaurant on Yelp. Because of this prevalence of choosing, there has been a resurgence of interest in choice models in the machine learning community. We explore new applications and stochastic models for choices, comparisons, and rankings in different contexts, and problems of large-scale inference and active learning for these models.
Can Who-Edits-What Predict Edit Survival? A. B. Yardım, V. Kristof, L. Maystre and M. Grossglauser 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, London, United Kingdom, August 19-23, 2018.
ChoiceRank: Identifying Preferences from Node Traffic in Networks L. Maystre and M. Grossglauser International Conference on Machine Learning, Sydney, Australia, August 6-11, 2017.
A large class of applications require to learn tasks on the fly and/or to (inter-) actively obtain new training samples from a data-source, for example when the environment is time-varying, or when it is computationally difficult to train off-line over the whole dataset because of its large size. We explore, develop and apply learning methods that balance the competing goal of gaining more knowledge by exploring new data, with that of optimizing its operation by exploiting the best configuration of the system based on current knowledge.
Wireless sensor and local networks rely on principles of self-organization for their operation. We develop algorithms to schedule and route transmissions in wireless networks and optimise throughput, delay, robustness and/or energy consumption. We also evaluate the benefits of combining different technologies such as wireless and power-line communications, which are widespread in home networking, to provide accrued and reliable performance in random and time-varying environments.
Multi-Armed Bandit in Action: Optimizing Performance in Dynamic Hybrid Networks S. Henri, C. Vlachou and P. Thiran IEEE/ACM Transactions on Networking, 2018-07-27.
EMPoWER Hybrid Networks: Exploiting Multiple Paths over Wireless and ElectRical Mediums S. Henri, C. Vlachou, J. Herzen and P. Thiran ACM Conference on emerging Networking EXperiments and Technologies (CoNEXT) 2016, Irvine, California, USA, December 12-15, 2016.
Electri-Fi Your Data: Measuring and Combining Power-Line Communications with WiFi C. Vlachou, S. Henri and P. Thiran ACM Internet Measurement Conference (IMC) 2015, Tokyo, Japan, October 28-30, 2015.