This position paper explores the challenges, existing solutions, and open issues related to resource allocation in federated learning environments. The focus is on how to allocate resources effectively while adhering to service level objectives (SLOs) and fairness requirements, which include factors such as server location, data provenance, energy consumption, sovereignty, carbon footprint, and economic cost. The goal is to optimise resource distribution across different stages of the federated learning process within a given architecture, ensuring that these fairness criteria are integrated into the allocation strategy. This approach aligns with decolonial methodologies that seek to offer more sustainable and equitable alternatives to the resource-intensive artificial intelligence processes prevalent today.

The omnipresence of services offered by diverse applications leads customers to share more and more personal data, among which some are sensitive. Dishonest entities perform inference attacks by querying non-sensitive data in order to deduce the stored sensitive data. Detecting those attacks is still an open problem in a setting where a dishonest entity has access to distinct data controllers' databases containing data collected from the same customer. This problem has been addressed considering a centralized detection system. However, this approach is limited because of this centralized nature where the system protects the customers' privacy at the expense of the data controllers' privacy. Hence, we propose in this article the description of a distributed architecture to detect inference attacks in a multi-database context, while preserving the privacy of both the applications and the customers.

Under the black-box model, data services do not export (meta)-data describing the conditions in which data are collected, in which they are deployed and processed, and the quality of the data they deliver. Thus, this model creates blind spots that prevent determining to which extent providers can be trusted to use their data services for building target applications. This paper proposes a QoS-based trust evaluation model for black box data services that combines QoS indicators, including service performance and data quality. The paper also introduces DETECT (Data sErvice as a black box Trust Evaluation arChitecTure) which validates our model. The experimental results demonstrate the feasibility and effectiveness of our solution.

Cloud provider barriers still stand. After a decade of cloud computing, customers struggle to overcome the challenge of crossing multi-provider clouds to benefit from fine-grained resource distribution, business independence from CSPs and cost savings. Although increasingly popular, most adopted IaaS intercloud solutions are generally limited to specific public cloud providers or present maintainability issues. Remaining hurdles include complexity of management and operations of such infrastructures, in presence of per-customer customizations and provider configurations. The Infrastructure as Code (IaC) paradigm is emerging as key enabler for IaaS multi-clouds, to develop and manage infrastructure configurations. However, due to complexity of the infrastructure life-cycle, to heterogeneity of composing resources and to user-customizations, this approach is far from being viable. In this paper, we explore an aspect-oriented approach to IaC deployment and management. We propose Mantus, a IaC-based multi-cloud builder composed of an aspect-oriented Domain-Specific Language called TML, or TOSCA Manipulation Language, and a corresponding aspect weaver to inject flexibly non-functional services in TOSCA infrastructure templates. We show the practical feasibility of our approach, with also good results in terms of performance and scalability.

Crowdsourcing has emerged as a promising approach for obtaining services and data in a short time and at a reasonable budget. However, the quality of the output provided by the crowd is not guaranteed, and must be controlled. This quality control usually relies on worker screening or contribution reviewing at the cost of additional time and budget overheads. In this paper, we propose to reduce these overheads by leveraging the system history. We describe an offline learning algorithm that groups tasks from history into homogeneous clusters and learns for each cluster the worker features that optimize the contribution quality. These features are then used by the online targeting algorithm to select reliable workers for each incoming task. The proposed method is compared to the state of the art selection methods using real world datasets. Results show that we achieve comparable, and in some cases better, output quality for a smaller budget and shorter time.

Opportunistic networks suffer some coarse characteristics as frequent disruptions, high loss ratios and uncertain data delivery. The transmission of video content even worsens the delivery problem because of the large size and continuous nature of the medium. Therefore, loss concealment is inducted as a part of the solution, alongside other delivery enhancement mechanisms. Loss concealment techniques are usually used Forwards, where the source must spend extra processing and transmission volume to enable the receiver to recover lost data parts. These techniques are not only associated with prolonged latencies and video encoding manipulations, but they are also sensitive to the loss pattern, which might leave the receiver in passive suspension if a key part for recovery was missing. In this work, we propose a Backward loss concealment mechanism, which allows the receiver to react to a certain amount of loss. Missing parts of the scalable video stream are compensated by a composite solution that encloses 1) video frame loss error concealment, and 2) network demands, which are initiated and cast in search for missing parts from other nodes in the network. Simulation-driven experiments prove the applicability of the backward mechanism, with a limited capability depending on the target application requirements.

Existing data integration techniques have to be revisited to query big data collections on the Cloud. Service Level Agreements implement the contracts between the cloud provider and the users, and between the cloud and service providers. Given SLA heterogeneity and data integration scalability problems, we propose an SLA guided data integration for querying data on multiple clouds.

Video transmission in opportunistic networks is very challenging due to the network coarse characteristics of frequent disruptions and uncertain data delivery. This is on the one hand, and the large size and continuous nature of videos on the other hand. In a previous work of Klaghstan et. al (2013), they presented a solution based on Scalable Video Coding (SVC), which aims at improving the user viewing experience assessed by means of playout delay and perceived quality. While this scheme obtained promising experimental results, it suffered from a non-negligible loss rate, which they explained by the still large size of the used SVC layers as the transmission unit. To address this issue, in this paper we investigate reducing the granularity of transmission units down to Network Abstraction Layer (NAL) units that compose SVC layers. The proposed solution packetizes NAL units into larger chunks in an adaptive way, dynamically taking into account environmental information gathered by each node (contact times with other nodes). Simulation-driven experimental results report a 100% playout availability and a significant improvement in the received quality level with no setback in delivery times.

Cloud computing is a revolutionary paradigm which enables on-demand provisioning of computing resources. Resources are delivered to cloud consumers in the form of infrastructure, platform and software services. These resources are deployed on three different models: private clouds, public clouds and hybrid clouds. In hybrid cloud context, private clouds externalize resources and invoke services from a public cloud when needed. However, in such a specific inter-cloud environment risks may arise. In fact, private cloud users often interact with cloud providers for services provisioning such as infrastructure, platforms and software. However, they are not sufficiently assured about how credible the data computed by these resources they have entrusted. This is due to clouds autonomy preservation, difference in control policy definitions and lack of transparency in clouds. In this paper, we propose a preventive/detective approach for assessing private cloud to select a trustworthy public cloud service. The solution is based on a mediator as a service that ensures the role of trust manager for the private cloud.

Opportunistic networks are generally characterized by a low performance due to user mobility and non end-to-end communications, which makes it challenging to use them in scenarios involving the transmission of large data, such as video transmission. This article presents simulated experiments of transmitting different video sequences in opportunistic networks, with the goal of enhancing user video viewing experience. We consider this subjective concept objectively as a trade-off between playout delay and the delivered quality level. To obtain the best trade-off, we propose using Scalable Video Coding, which divides a video into multiple unequally important layers, providing different levels of quality. This property is exploited to transmit video layers with unequal degrees of redundancy proportionally to their degree of importance. Experimental results show that this approach enables the video to become more quickly playable at the destination with a low starting quality but progressively being enhanced, providing a better viewing-experience.

The vision of Universal Multimedia Access (UMA) and Universal Multimedia Experience (UME) has driven research in the multimedia community for a long time. Implementing content adaptation frameworks to satisfy heterogeneous types of constraints is among the main requirements for UMA and UME. At the core of these frameworks lies the adaptation decision engine, which computes the appropriate adaptation plans. Though much research has already been done in this domain, the problem of semantic constraints has generally been neglected. This paper addresses the problem of selecting the optimal adaptation operation to satisfy semantic constraints while maximizing the utility of the adapted video. To this end, we define a utility function that computes a value for each possible adaptation operation. We represent our utility function using the MPEG-21 Digital Item Adaptation (DIA) tools. This facilitates the integration of semantic constraints with other types of constraints in MPEG-21 Universal Constraint Description format.

Data replication can improve data availability in mobile networks; but due to typical resource limitations in such environments, specific replication mechanisms are needed. In this paper, we propose CReaM, user-Centric REplicAtion Model for mobile environment. This model puts users at the centre by letting them determine the amount of resources they are willing to share. CReaM is suitable for dynamic environments where each node needs a certain level of decision autonomy. In this paper, we present CReaM's general principles and focus on one of its core function, which is its autonomic behavior that generates replication requests based on resources monitoring and user settings.

In the last years, several distributed system paradigms have emerged, aiming at the share of a large amount of resources among a large number of users. The virtual organization concept in a multi-institutional context obviously presents numerous opportunities for business applications as well as for scientific ones. Nevertheless, the current trends seem to lead to several independent specialized grids in opposition to the early vision of one generic world wide grid. In such a context, large scale heterogeneous databases could be seen as several specialized databases hosted by different grids. The challenge is to allow users of these grids to access all databases with the same efficiency as if they were supported by the same grid. Our proposal is an overlay software architecture that is built on top of grid middlewares and acts in behalf of the user to facilitate the use of inter-grid databases. The proposed mediation solution deals with database heterogeneity and access rights management. It is capable of making decisions regarding the selection of resources and according to the hosting grid and to the infrastructure performances.

The recent emergence of handheld devices and wireless networks has implied an exponential increase of terminals users. So, today, service providers have to propose new applications adapted to mobile environments. In this paper, we propose and describe a new class of distributed applications called the proximity applications. In such applications, two or more handheld devices, physically close to each other, can communicate and exchange data in a secure way. Proximity applications rely on the use of both different mobile devices and heterogeneous wireless networks. Thus, these applications need a high degree of flexibility, for an easy and rapid application development. In this context, our purpose is to study the interest of the Hybrid Peer-To-Peer (P2P) architecture model use specially for the extensibility, the fault-tolerance management and the scalability of proximity applications. Moreover, thanks to this model, proximity applications can easily face to the heterogeneity of devices and networks.