Exploit (informatique) | EPFL Graph Search

Un exploit ou code d'exploitation est, dans le domaine de la sécurité informatique, un élément de programme permettant à un individu ou à un logiciel malveillant d'exploiter une faille de sécurité informatique dans un système informatique. Que ce soit à distance (remote exploit) ou sur la machine sur laquelle cet exploit est exécuté (local exploit), le but de cette manœuvre est de s'emparer des ressources d'un ordinateur ou d'un réseau, d'accroître le privilège d'un logiciel ou d'un utilisateur sur la machine-cible, ou encore d'effectuer une attaque par déni de service. Les vers exploitent des vulnérabilités distantes pour se diffuser automatiquement d'un PC à un autre au sein d'un réseau. Par exemple, le ver Conficker, pour se diffuser, exploite la vulnérabilité (MS08-067) au sein du système d'exploitation Windows. Étymologie Le terme « exploit » provient de l'anglais exploit, pour « exploiter » (ici, une faille de sécurité). L'usage est de le prononcer à l'anglaise o

A Federated Recommender System for Online Learning Environments

Denis Gillet, Laurent Moccozet, Christophe Salzmann, Lei Zhou

From e-commerce to social networking sites, recommender systems are gaining more and more interest. They provide connections, news, resources, or products of interest. This paper presents a federated recommender system, which exploits data from different online learning platforms and delivers personalized recommendation. The underlying educational objective is to enable academic institutions to provide a Web 2.0 dashboard bringing together open resources from the Cloud and proprietary content from in-house learning management systems. The paper describes the main aspects of the federated recommender system, including its adopted architecture, the common data model used to harvest the different learning platforms, the recommendation algorithm, as well as the recommendation display widget.

Springer Berlin Heidelberg2012

Measuring and Managing Switch Diversity in Software Defined Networks

Maciej Leszek Kuzniar

Software Defined Networking (SDN) is a novel approach to building computer networks. Improved flexibility, simplified management and cost reduction promised by SDN makes many see it as the future of networking. The main insight of SDN is the separation of network control and configuration decisions from packet forwarding devices. This way, complex control software development becomes independent of changes in hardware traffic forwarding switches. Relying on a well-defined interface, a controller program can supervise the whole network built of many switches, produced by multiple vendors. I argue however, that this idealized vision is difficult to achieve in practice and that an SDN controller cannot treat all switches equally. There are multiple reasons why switches following the same specification are heterogeneous: unclear specification, implementation difficulties, cost, and human errors. In this dissertation, I describe a two-phase approach to handle switch diversity in SDN. First, I present systematic techniques to detect various differences between devices. SOFT is a method and a tool that uncovers functional differences in switch software. It relies on established software analysis techniques and a novel application of a constraint solver to find inputs that result in distinct behavior of two switches. Further, I design a systematic methodology for switch update performance measurements. A benchmarking tool based on this technique computes update rate characteristics of hardware switches. By additionally tracking interactions between control and data planes during the update, this methodology is capable of revealing inconsistencies showing that the switches do not follow the specification. Finally, I address the most severe issue detected. RUM, a software layer between the controller and the switches masks and fixes incorrect rule update notifications coming from faulty switches. I demonstrate the usefulness of the described solutions by analyzing existing switches. SOFT detected several inconsistencies between two open source software switches. The performance benchmark revealed errors that compromise network security. It also provided detailed switch characteristics that should be taken into account by controller developers to improve network performance. RUM prevents packet drops in a safe network update without requiring any changes to faulty switches.

EPFL2016

Open-Ended Learning of Visual and Multi-Modal Patterns

Jie Luo

A common trend in machine learning and pattern classification research is the exploitation of massive amounts of information in order to achieve an increase in performance. In particular, learning from huge collections of data obtained from the web, and using multiple features generated from different sources, have led to significantly boost of performance on problems that have been considered very hard for several years. In this thesis, we present two ways of using these information to build learning systems with robust performance and some degrees of autonomy. These ways are Cue Integration and Cue Exploitation, and constitute the two building blocks of this thesis. In the first block, we introduce several algorithms to answer the research question on how to integrate optimally multiple features. We first present a simple online learning framework which is a wrapper algorithm based on the high-level integration approach in the cue integration literature. It can be implemented with existing online learning algorithms, and preserves the theoretical properties of the algorithms being used. We then extend the Multiple Kernel Learning (MKL) framework, where each feature is converted into a kernel and the system learns the cue integration classifier by solving a joint optimization problem. To make the problem practical, We have designed two new regularization functions making it possible to optimize the problem efficiently. This results in the first online method for MKL. We also show two algorithms to solve the batch problem of MKL. Both of them have a guaranteed convergence rate. These approaches achieve state-of-the-art performance on several standard benchmark datasets, and are order of magnitude faster than other MKL solvers. In the second block, We present two examples on how to exploit information between different sources, in order to reduce the effort of labeling a large amount of training data. The first example is an algorithm to learn from partially annotated data, where each data point is tagged with a few possible labels. We show that it is possible to train a face classification system from data gathered from Internet, without any human labeling, but generating in an automatic way possible lists of labels from the captions of the images. Another example is under the transfer learning setting. The system uses existing models from potentially correlated tasks as experts, and transfers their outputs over the new incoming samples, of a new learning task where very few labeled data are available, to boost the performance.