Fermer
_fs_informatique_choix.jpg

Projets de Master

Plusieurs projets de M.Sc. peuvent être faits à l’institut pour les étudiants inscrits au Swiss Joint Master in Computer Science. Quelques projets sont décrits ci-bas, mais n’hésitez pas à nous contacter si vous avez un autre sujet de projet qui vous intéresse.

Pour plus d’information à propos du Swiss Joint Master in Computer Science, veuillez visiter le lien suivant : 

SafeCloud - general

Contact: 
Office B121
 
Office B114
 
Cloud infrastructures, despite all their advantages and importance to the competitiveness of modern economies, raise fundamental questions related to the privacy, integrity, and security of offsite data storage and processing tasks. These questions are currently not answered satisfactorily by existing technologies. Furthermore, recent developments in the wake of the expansive and sometimes unauthorised government access to private and sensitive data raise major privacy and security concerns about data located in the cloud, especially when data is physically located, processed, or must transit outside the legal jurisdiction of its rightful owner. This is exacerbated by providers of cloud services that frequently move and process data without notice in ways that are detrimental to the users and their privacy.
 
SafeCloud, a consortium of seven research and industrial partners, will re-architect cloud infrastructures to ensure that data transmission, storage, and processing can be (1) partitioned in multiple administrative domains that are unlikely to collude, so that sensitive data can be protected by design; (2) entangled with inter-dependencies that make it impossible for any of the domains to tamper with its integrity. These two principles (partitioning and entanglement) are thus applied holistically across the entire data management stack, from communication to storage and processing.
 
There are several practical and theoretical MSc projects covering all the aspects of SafeCloud (coding for storage devices, security, system integration, etc.). Contact us if some aspects of the project might interest you.
 

 

SafeCloud - Erasure codes for distributed storage systems

Contact: 
Dr Hugues Mercier
Office B121

Distributed file storage, sharing and synchronization systems on the cloud are becoming ubiquitous, and all major players are offering some flavor of it: Dropbox, iCloud, Google Drive, OneDrive, Amazon S3, ... They provide easy access to data from multiple devices as well as protection against data loss from hardware failures. Protecting such data against hardware and link failures is a very active area of theoretical and applied research. It is the topic of this project.

As part of the SafeCloud project, we want to study and implement the state-of-the-art storage error-correcting codes within the SafeCloud platform, and how these codes can support its privacy and security features.

http://www.safecloud-project.eu/


SafeCloud - security

Contact: 
Dr Hugues Mercier
Office B121

SafeCloud, a consortium of seven research and industrial partners, will re-architect cloud infrastructures to ensure that data transmission, storage, and processing can be (1) partitioned in multiple administrative domains that are unlikely to collude, so that sensitive data can be protected by design; (2) entangled with inter-dependencies that make it impossible for any of the domains to tamper with its integrity. These two principles (partitioning and entanglement) are thus applied holistically across the entire data management stack, from communication to storage and processing.

As part of SafeCloud, we are looking for Master students to study various privacy and security aspects of the platform. This includes plausible-deniability, proofs of storage and proofs of redundancy. The objective is to provide verifiable guarantees to users that their data is properly, securely, and reliably archived. For instance, if a storage provider guarantees that the equivalent of three copies of each piece of data is archived on three continents, how can users verify that this claim is more than a marketing slogan? In practice, it appears difficult to prove this claim in a simple and convincing way. Users rely mostly in the good faith they have in their providers (and in the catastrophic consequences for their providers’ bottom line should they lose the data). 

http://www.safecloud-project.eu/


Balls-and-bins probabilistic protocols for ordering of events in distributed systems

Contact: 
Dr Hugues Mercier
Office B121

The ordering of events is one of the most fundamental and studied problems in distributed systems, and until recently, the main research focus was on the construction of primitives with strong deterministic guarantees. However, the impressive growth of large-scale distributed systems exposed the practical weaknesses of these approaches: poor scalability, unacceptable degraded behaviour under churn, and increased latency. These practical weaknesses led to the development of alternative formulations with weaker yet quantifiable guarantees such as epidemic (probabilistic) dissemination protocols.

Probabilistic protocols have many advantages, and give rise to many theoretical and applied mathematical problems that can be modelled with a balls-and-bins model. A balls-and-bins model abstracts processes as bins and messages as balls and studies, for instance, how many balls need to be thrown such that each bin receives at least a ball with arbitrarily high probability given various constraints (number of dissemination rounds, churn, message loss, latency, bandwidth, ...). The existing asymptotic bounds in the literature are loose, and as a result the algorithms are significantly more costly than they could be. This problem exhibits interesting similarities with biometrics problems studied in the 1960s.

This project is in collaboration with University of Minho in Portugal and will potentially lead to the development of improved very large scale data stores. I would gladly supervise or co-supervise a Mathematics or CS student who would like to do his Master thesis on the mathematical and computational aspects of this project. 

 

[1] Miguel Matos, Hugues Mercier, Pascal Felber, Rui Oliveira, and José Pereira. EpTO: an Epidemic Total Order Algorithm for Large-Scale Distributed Systems. 2015 ACM/IFIP/USENIX Middleware conference. 


 

Low complexity decoding of variable-length codes for video compression

Contact: 
Dr Hugues Mercier
Office B121

Latest video compression standards [1] achieve compression by removing redundancy from the source symbols, however this makes them very sensitive to channel noise [2]. Most current video compression schemes use variable length codes, but classic decoding algorithms like maximum a posteriori decoding are costly due to the lack of synchronization between symbol sequences and coded bit sequences, resulting in algorithms with prohibitive complexity. The problem of reducing this high complexity has been studied for more than 15 years [3] and is the topic of this project.

The first objective of the project is to survey the different techniques used to reduce the com- plexity of current algorithms for decoding variable length source-channel codes. The second objective is to compare the existing techniques with a novel technique [5] by running simulations with state-of-the-art video sequences.

[1] D. Marpe, T. Wiegand, and G. Sullivan. The H.264/MPEG–4 advanced video coding standard and its applications. IEEE Communications Magazine, vol. 44, no. 8, pp. 134–143, 2006.

[2] P. Duhamel and M. Kieffer. Joint Source-Channel Decoding: A Cross-Layer Perspective with Applications in Video Broadcasting over Mobile and Wireless Networks. pp. 123–128, EURASIP and Academic Press Series in Signal and Image Processing, 2010.

[3] G. Mohammad-Khani, C. Lee, M. Kieffer, and P. Duhamel. Simplification of VLC tables with application to ML and MAP decoding algorithms. IEEE Transactions on Communications, vol. 54, no. 10, pp. 1835–1844, 2006.

[5] A. Wasae and H. Mercier and F. Labeau. Low complexity decoding of variable length source-channel codes. 39th International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2014.