The ever-increasing interest in and widespread diffusion of Machine Learning (ML)-based applications has driven a substantial amount of research into offensive and defensive ML. ML models can be attacked from different angles: poisoning attacks, the focus of this paper, inject maliciously crafted data points in the training set to modify the model behavior; adversarial attacks maliciously manipulate inference-time data points to fool the ML model and drive the prediction of the ML model according to the attacker’s objective. Ensemble-based techniques are among the most relevant defenses against poisoning attacks and replace the monolithic ML model with an ensemble of ML models trained on different (disjoint) subsets of the training set. They assign data points to the training sets of the models in the ensemble (routing) randomly or using a hash function, assuming that evenly distributing poisoned data points positively influences ML robustness. Our paper departs from this assumption and implements a risk-based ensemble technique where a risk management process is used to perform a smart routing of data points to the training sets. An extensive experimental evaluation demonstrates the effectiveness of the proposed approach in terms of its soundness, robustness, and performance.

Modern applications are increasingly driven by Machine Learning (ML) models whose non-deterministic behavior is affecting the entire application life cycle from design to operation. The pervasive adoption of ML is urgently calling for approaches that guarantee a stable non-functional behavior of ML-based applications over time and across model changes. To this aim, non-functional properties of ML models, such as privacy, confidentiality, fairness, and explainability, must be monitored, verified, and maintained. Existing approaches mostly focus oni)implementing solutions for classifier selection according to the functional behavior of ML models,ii)finding new algorithmic solutions, such as continuous re-training. In this paper, we propose a multi-model approach that aims to guarantee a stable non-functional behavior of ML-based applications. An architectural and methodological approach is provided to compare multiple ML models showing similar non-functional properties and select the model supporting stable non-functional behavior over time according to (dynamic and unpredictable) contextual changes. Our approach goes beyond the state of the art by providing a solution that continuously guarantees a stable non-functional behavior of ML-based applications, is ML algorithm-agnostic, and is driven by non-functional properties assessed on the ML models themselves. It consists of a two-step process working during application operation, wheremodel assessmentverifies non-functional properties of ML models trained and selected at development time, andmodel substitutionguarantees continuous and stable support of non-functional properties. We experimentally evaluate our solution in a real-world scenario focusing on non-functional property fairness.

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.

Modern service-based systems are characterized by applications composed of heterogeneous services provided by multiple, untrusted providers, and deployed along the (multi-) cloud-edge continuum. This scenario of increasing pervasiveness, complexity, and multi-party service recruitment urgently calls for solutions to increase applications privacy and security, on the one hand, and guarantee that applications behave as expected and support a given set of non-functional requirements, on the other hand. Certification schemes became the widespread means to answer this call, but they still build on old-fashioned assumptions that hardly hold in today’s services world. They assume that all actors involved in a certification process are trusted "by definition", meaning that certificates are supposed to be correct and be safely usable for decision-making, such as certification-based service selection and composition. In this paper, we depart from such unrealistic assumptions and define the first certification scheme that is completely transparent to the involved actors and significantly more resistant to misbehavior (e.g., collusion). We design a blockchain-based architecture to support our scheme, redefining the actors and their roles. The quality and performance of our scheme are evaluated in a case study scenario.

The huge progress of ICT is radically changing distributed systems at their roots, modifying their operation and engineering practices and introducing new non-functional (e.g., security and safety) risks. These risks are amplified by the crucial role played by machine learning, on one side, and by the pervasive involvement of users in the system operation, on the other side. Certification techniques have been largely adopted to reduce the above risks, though the recent evolution of distributed systems towards cloud-edge, IoT, 5G, and machine learning severely hindered certification diffusion and quality. The need of new certification techniques that prove compliance of distributed systems against non-functional requirements arises and is often pushed by strict laws and regulations. In this paper, we envision a research manifesto for non-functional certification of modern distributed systems that paves the way for the wide adoption of certification in the real world, also in those domains where certification is not mandatory. Its ultimate goal is to lead to a trustworthy and adaptive ecosystem based on a cost-effective, non-functional certification, where modern system development, assessment, and management are not only ruled by functional requirements. The manifesto discusses the research challenges, a roadmap built on 6 research directions, and a concrete implementation timeline for the roadmap.

The cloud computing has deeply changed how distributed systems are engineered, leading to the proliferation of ever/evolving and complex environments, where legacy systems, microservices, and nanoservices coexist. These services can severely impact on individuals' security and safety, introducing the need of solutions that properly assess and verify their correct behavior. Security assurance stands out as the way to address such pressing needs, with certification techniques being used to certify that a given service holds some non/functional properties. However, existing techniques build their evaluation on software artifacts only, falling short in providing a thorough evaluation of the non/functional properties under certification. In this paper, we present a multi/dimensional certification scheme where additional dimensions model relevant aspects (e.g., programming languages and development processes) that significantly contribute to the quality of the certification results. Our multi/dimensional certification enables a new generation of service selection approaches capable to handle a variety of user's requirements on the full system life cycle, from system development to its operation and maintenance. The performance and the quality of our approach are thoroughly evaluated in several experiments.

Machine learning is becoming ubiquitous. From finance to medicine, machine learning models are boosting decision/making processes and even outperforming humans in some tasks. This huge progress in terms of prediction quality does not however find a counterpart in the security of such models and corresponding predictions, where perturbations of fractions of the training set (poisoning) can seriously undermine the model accuracy. Research on poisoning attacks and defenses received increasing attention in the last decade, leading to several promising solutions aiming to increase the robustness of machine learning. Among them, ensemble-based defenses, where different models are trained on portions of the training set and their predictions are then aggregated, provide strong theoretical guarantees at the price of a linear overhead. Surprisingly, ensemble-based defenses, which do not pose any restrictions on the base model, have not been applied to increase the robustness of random forest models. The work in this paper aims to fill in this gap by designing and implementing a novel hash-based ensemble approach that protects random forest against untargeted, random poisoning attacks. An extensive experimental evaluation measures the performance of our approach against a variety of attacks, as well as its sustainability in terms of resource consumption and performance, and compares it with a traditional monolithic model based on random forest. A final discussion presents our main findings and compares our approach with existing poisoning defenses targeting random forests.

While certification is widely recognized as a means to increase system trustworthiness and reduce uncertainty in decision making, it faces severe challenges preventing a wider adoption thereof. Certification is not adequately planned and integrated within the development process, leading to suboptimal scenarios where certification introduces the need to further modify the developed system with high costs. We propose a methodology that bridges the gap between software development and certification processes. Our methodology automatically produces the certification requirements driving all steps of the development process, and maximizes the strength of certificates while taking costs under control. We formalize the above problem as a multi-objective mathematical program and solve it through a genetic algorithm. The proposed approach is tested in a real-world, cloud-based financial scenario at CaixaBank and its performance and quality is evaluated in a simulated scenario.

Today big data pipelines are increasingly adopted by service applications representing a key enabler for enterprises to compete in the global market. However, the management of non-functional aspects of the big data pipeline (e.g., security, privacy) is still in its infancy. As a consequence, while functionally appealing, the big data pipeline does not provide a transparent environment, impairing the users’ ability to evaluate its behavior. In this paper, we propose a security assurance methodology for big data pipelines grounded on the DevSecOps development paradigm to increase trustworthiness allowing reliable security and privacy by design. Our methodology models and annotates big data pipelines with non-functional requirements verified by assurance checks ensuring requirements to hold along with the pipeline lifecycle. The performance and quality of our methodology are evaluated in a real walkthrough analytics scenario.

Modern distributed systems consist of a multilayer architecture of IoT, edge, and cloud nodes. Together, they are revolutionizing our lives, bringing intelligence to existing processes (e.g., smart grids) and enabling novel, efficient and effective processes (e.g., remote surgery). This transition however does not come without drawbacks, due to the ever-increasing reliance on devices whose security and safety are, at least, questionable. In this context, research is in its infancy, struggling to adapt successful practices applied, for instance, in cloud systems. Security of modern IoT systems still relies on old-fashioned approaches, mostly static assessments considering only very specific parts of the target system, rather than assessing the system as a whole. In this paper, we put forward the idea of security assurance for IoT, as a higher-level assurance process evaluating the target system at different layers and different moments of its lifecycle, then implemented by a flexible assurance framework. The quality of our approach is evaluated in a real-world smart lighting system.

The advent of cloud computing and Internet of Things (IoT) has deeply changed the design and operation of IT systems, affecting mature concepts like trust, security, and privacy. The benefits in terms of new services and applications come at a price of new fundamental risks, and the need of adapting risk management frameworks to properly understand and address them. While research on risk management is an established practice that dates back to the 90s, many of the existing frameworks do not even come close to address the intrinsic complexity and heterogeneity of modern systems. They rather target static environments and monolithic systems thus undermining their usefulness in real-world use cases. In this paper, we present an assurance-based risk management framework that addresses the requirements of risk management in modern distributed systems. The proposed framework implements a risk management process integrated with assurance techniques. Assurance techniques monitor the correct behavior of the target system, that is, the correct working of the mechanisms implemented by the organization to mitigate the risk. Flow networks compute risk mitigation and retrieve the residual risk for the organization. The performance and quality of the framework are evaluated in a simulated industry 4.0 scenario.