Guillaume Rosinosky

Bio: Guillaume Rosinosky is an academic researcher from Université catholique de Louvain. The author has contributed to research in topics: Software as a service & Legacy system. The author has co-authored 3 publications.

PProx: efficient privacy for recommendation-as-a-service

Abstract: We present PProx, a system preventing recommendation-as-a-service (RaaS) providers from accessing sensitive data about the users of applications leveraging their services. PProx does not impact recommendations accuracy, is compatible with arbitrary recommendation algorithms, and has minimal deployment requirements. Its design combines two proxying layers directly running inside SGX enclaves at the RaaS provider side. These layers transparently pseudonymize users and items and hide links between the two, and PProx privacy guarantees are robust even to the corruption of one of these enclaves. We integrated PProx with Harness's Universal Recommender and evaluated it on a 27-node cluster. Our results indicate its ability to withstand a high number of requests with low end-to-end latency, horizontally scaling up to match increasing workloads of recommendations.

A Methodology for Tenant Migration in Legacy Shared-Table Multi-tenant Applications

Abstract: Multi-tenancy enables cost-effective SaaS through resource consolidation. Multiple customers, or tenants, are served by a single application instance, and isolation is enforced at the application level. Service load for different tenants can vary over time, requiring applications to scale in and out. A large class of SaaS providers operates legacy applications structured around a relational (SQL) database. These applications achieve tenant isolation through dedicated fields in their relational schema and are not designed to support scaling operations. We present a novel solution for scaling in or out such applications through the migration of a tenant’s data to new application and database instances. Our solution requires no change to the application and incurs no service downtime for non-migrated tenants. It leverages external tables and foreign data wrappers, as supported by major relational databases. We evaluate the approach using two multi-tenant applications: Iomad, an extension of the Moodle Learning Management System, and Camunda, a business process management platform. Our results show the usability of the method, minimally impacting performance for other tenants during migration and leading to increased service capacity after migration.

Optimisation of business process tenant distribution in the Cloud with a genetic algorithm

Abstract: With the generalization of the Cloud, software providers can distribute their software as a service without investing in large infrastructure. However, without an effective resource allocation method, their operation cost can grow quickly, hindering the profitability of the service. This is the case for BPM as a Service providers that want to handle hundreds of customers with a given quality of service. Since there are variations in the capacity and the number of users, the allocation method must be able to adjust the resource and the allocation of customer on these resources. In this paper we present a cost optimization model and a heuristic based on genetic algorithms to adjust resource allocation to the need of a set of customers with varying BPM task throughput. Experi-mentations using realistic customer loads and cloud resources capacities shows the gain of this method compared to previous approaches.

Privacy-preserving techniques in recommender systems: state-of-the-art review and future research agenda

Abstract: PurposeThis study explores privacy challenges in recommender systems (RSs) and how they have leveraged privacy-preserving technology for risk mitigation. The study also elucidates the extent of adopting privacy-preserving RSs and postulates the future direction of research in RS security.Design/methodology/approachThe study gathered articles from well-known databases such as SCOPUS, Web of Science and Google scholar. A systematic literature review using PRISMA was carried out on the 41 papers that are shortlisted for study. Two research questions were framed to carry out the review.FindingsIt is evident from this study that privacy issues in the RS have been addressed with various techniques. However, many more challenges are expected while leveraging technology advancements for fine-tuning recommenders, and a research agenda has been devised by postulating future directions.Originality/valueThe study unveils a new comprehensive perspective regarding privacy preservation in recommenders. There is no promising study found that gathers techniques used for privacy protection. The study summarizes the research agenda, and it will be a good reference article for those who develop privacy-preserving RSs.

Community Detection Attack against Collaborative Learning-based Recommender Systems

Abstract: Collaborative-learning based recommender systems emerged following the success of collaborative learning techniques such as Federated Learning (FL) and Gossip Learning (GL). In these systems, users participate in the training of a recommender system while keeping their history of consumed items on their devices. While these solutions seemed appealing for preserving the privacy of the participants at a first glance, recent studies have shown that collaborative learning can be vulnerable to a variety of privacy attacks. In this paper we propose a novel privacy attack called Community Detection Attack (CDA), which allows an adversary to discover the members of a community based on a set of items of her choice (e.g., discovering users interested in LGBT content). Through experiments on three real recommendation datasets and by using two state-of-the-art recommendation models, we assess the sensitivity of an FL-based recommender system as well as two flavors of Gossip Learning-based recommender systems to CDA. Results show that on all models and all datasets, the FL setting is more vulnerable to CDA than Gossip settings. We further evaluated two off-the-shelf mitigation strategies, namely differential privacy (DP) and a share less policy, which consists in sharing a subset of model parameters. Results show a better privacy-utility trade-off for the share less policy compared to DP especially in the Gossip setting.

Efficient privacy preserving top-k recommendation using homomorphic sorting

Abstract: . Theexistingworksonprivacy-preservingrecommendersystemsbased on homomorphic encryption do not filter top-k most relevant items on the server side. As a result, sending the encrypted rating vector for all items to the user retrieving the top-k items is necessary. This incurs significant computation and communication costs on the user side. In this work, we employ private sorting at the server to reduce the user-side overheads. In private sorting, the values and corresponding positions of elements must remain private. We use an existing private sorting protocol by Foteini and Olga and tailor it to the privacy-preserving top-k recommendation applications. We enhance it to use secure bit decomposition in the private comparison routine of the protocol. This leads to a notable reduction in cost overheads of users as well as the servers, especially at the keyserver where the computation cost is reduced to half. The dataserver does not have to perform costly encryption and decryption operations. It performs computationally less expensive modular exponentiation operations. Since the private comparison operation contributes significantly to the overall cost overhead, making it efficient enhances the sorting protocol’s performance. Our security analysis con-cludes that the proposed scheme is as secure as the original protocol.