Luís Rodrigues

With the growing interest in blockchains, permissioned approaches to consensus have received increasing attention. Unfortunately, the BFT consensus algorithms that are the backbone of most of these blockchains scale poorly and offer limited throughput. In fact, many state-of-the-art BFT consensus algorithms require a single leader process to receive and validate votes from a quorum of processes and then broadcast the result, which is inherently non-scalable. Recent approaches avoid this bottleneck by using dissemination/aggregation trees to propagate values and collect and validate votes. However, the use of trees increases the round latency, which limits the throughput for deeper trees. In this article, we propose Kauri, a BFT communication abstraction that sustains high throughput as the system size grows by leveraging a novel pipelining technique to perform scalable dissemination and aggregation on trees. Furthermore, when the number of faults is moderate (arguably the most common case in practice), our construction is able to recover from faults in an optimal number of reconfiguration steps. We implemented and experimentally evaluated Kauri with up to 800 processes. Our results show that Kauri outperforms the throughput of state-of-the-art permissioned blockchain protocols, by up to 58x without compromising latency. Interestingly, in some cases, the parallelization provided by Kauri can also decrease the latency.

Machine Learning (ML)-enabled systems are now ubiquitous. These systems leverage ML models that require frequent re-trainings, which can lead to unsustainable costs. To prevent superfluous retrains and optimize system performance in the long run, we introduce Ripple, a system that leverages probabilistic model checking techniques to automate the decision of when to retrain the ML models employed by an ML-enabled system. The main challenge tackled by Ripple is estimating how the predictive quality of an ML model varies over long-term horizons, both when it is and when it is not retrained. Ripple introduces Look-Ahead Adaptation Impact Predictors (LA-AIPs) that are used in combination with a probabilistic model checker to determine whether to retrain the ML model. This allows Ripple to optimize system performance in the long term, contributing to creating more sustainable ML-enabled systems. We demonstrate Ripple’s feasibility via a fraud detection system use case, showcasing its ability to plan for the long term and to account for retrain latency, improving over myopic adaptation approaches.

This paper proposes a new algorithm for collision-free coverage control of multiple non-cooperating swarms in the presence of bounded disturbances. A new methodology is introduced that accounts for uncertainties in disturbance measurements. The proposed methodology is used to develop an algorithm that ensures collision-free motion in multi-swarm coverage control, specifically for cases where disturbances are present and their measurements are subject to bounded uncertainty. The theoretical results are validated through simulations of multiple swarms that independently aim to cover a given region in an environment with disturbances.

This paper formulates sufficient conditions for observability of a class of linear time invariant systems with a norm output. The observability of these systems depends on the input and will be studied for any input that satisfies a linear autonomous difference equation. It is shown that the observability conditions are related to the navigation problem of finding a position fix from several range measurements by intersecting spheres. This offers a new geometric interpretation of observability. Moreover, it will be shown that the initial condition of the system can still be estimated in the least squares sense if noise is added to the output. An example shows the application of the methodology. Additionally the example shows that the initial condition of the noiseless system is obtained by inverting a lower dimensional upper triangular matrix. Thus, the proposed method yields a matrix inversion with less computations when compared to the augmented state linear time-varying system approaches in the literature. A second example shows how the proposed methodology can be extended to perform state estimation.

The use of remote storage has become prevalent both by organizations and individuals. By relying on third-party storage, such as cloud or peer-to-peer storage services, availability, fault tolerance, and low access latency can be attained in a cost-efficient manner. Unfortunately, storage providers may misbehave and violate Service-Level Agreements (SLAs). In this article, we propose a new Proof of Timely-Retrievability (PoTR) that aims at assessing whether a provider is able to retrieve data objects with a latency lower than some SLA-specific threshold δ. We have implemented the PoTR and evaluated two distinct configurations of the proof, one tailored to estimate the average latency experience by clients and the other tailored to assess its variance. We leverage Trusted Execution Environments (e.g., Intel SGX) to ensure that the proof is produced by the node being audited and to reduce the communication between the auditor and the audited node. We have experimentally evaluated our prototypes considering a challenging edge computing setting, where storage services are provided by resource-constrained fog nodes, and the distance between the auditor and the audited node can be large. Despite the noise introduced by edge network delays, we show that the auditor is able to effectively detect SLA violations.

This paper provides a least squares formulation for the training of a 2-layer convolutional neural network using quadratic activation functions, a 2-norm loss function, and no regularization term. Using this method, an analytic expression for the globally optimal weights is obtained alongside a quadratic input-output equation for the network. These properties make the network a viable tool in system theory by enabling further analysis, such as the sensitivity of the output to perturbations in the input, which is crucial for safety-critical systems such as aircraft or autonomous vehicles. The least squares method is compared to previously proposed strategies for training quadratic networks and to a back-propagation-trained ReLU network. The proposed method is applied to a system identification problem and a GPS position estimation problem. The least squares network is shown to have a significantly reduced training time with minimal compromises on prediction accuracy alongside the advantages of having an analytic input-output equation. Although these results only apply to 2-layer networks, this paper motivates the exploration of deeper quadratic networks in the context of system theory.

This article focuses on the problem of optimizing the system utility of Machine Learning (ML)-based systems in the presence of ML mispredictions. This is achieved via the use of self-adaptive systems and through the execution of adaptation tactics, such asmodel retraining, which operate at the level of individual ML components.To address this problem, we propose a probabilistic modeling framework that reasons about the cost/benefit tradeoffs associated with adapting ML components. The key idea of the proposed approach is to decouple the problems of estimating (1) the expected performance improvement after adaptation and (2) the impact of ML adaptation on overall system utility.We apply the proposed framework to engineer a self-adaptive ML-based fraud detection system, which we evaluate using a publicly available, real fraud detection dataset. We initially consider a scenario in which information on the model’s quality is immediately available. Next, we relax this assumption by integrating (and extending) state-of-the-art techniques for estimating the model’s quality in the proposed framework. We show that by predicting the system utility stemming from retraining an ML component, the probabilistic model checker can generate adaptation strategies that are significantly closer to the optimal, as compared against baselines such as periodic or reactive retraining.