Search

Scenario generation is a crucial task in Stochastic Programming (SP). It involves a trade-off between keeping the scenario set small while making it representative for the target problem. While most state-of-the-art methods focus on matching the uncertainty of the stochastic process using distribution-driven approaches, problem-driven methodologies have been proposed in recent years to exploit the structure of the target problem during the scenario generation process. In order to represent uncertainties in a more concise way, we propose a novel approach based on Active Learning that sequentially generates a set of scenarios by including a new promising scenario at each iteration. Searching for the most promising scenario is a black-box global optimization problem, efficiently solved via Bayesian Optimization. Preliminary experimental results are presented on a classical newsvendor problem, providing empirical evidence that the proposed method can both identify the smallest and most informative scenario set for the problem. Our method can also efficiently and effectively handle multi-modal and fat-tailed distributions, analogously to the most recent problem-driven methods.

We introduce QuaterGCN, a spectral Graph Convolutional Network (GCN) with quaternion-valued weights at whose core lies the Quaternionic Laplacian, a quaternion-valued Laplacian matrix by whose proposal we generalize two widely-used Laplacian matrices: the classical Laplacian (defined for undirected graphs) and the complex-valued Sign-Magnetic Laplacian (proposed to handle digraphs with weights of arbitrary sign). In addition to its generality, our Quaternionic Laplacian is the only Laplacian to completely preserve the topology of a digraph, as it can handle graphs and digraphs containing antiparallel pairs of edges (digons) of different weights without reducing them to a single (directed or undirected) edge as done with other Laplacians. Experimental results show the superior performance of QuaterGCN compared to other state-of-the-art GCNs, particularly in scenarios where the information the digons carry is crucial to successfully address the task at hand.

Many real-life systems are usually controlled through policies replicating experts’ knowledge, typically favouring “safety” at the expense of optimality. Indeed, these control policies are usually aimed at avoiding a system’s disruptions or deviations from a target behaviour, leading to suboptimal performances. This paper proposes a statistical learning approach to exploit the historical safe experience—collected through the application of a safe control policy based on experts’ knowledge— to “safely explore” new and more efficient policies. The basic idea is that performances can be improved by facing a reasonable and quantifiable risk in terms of safety. The proposed approach relies on Gaussian Process regression to obtain a probabilistic model of both a system’s dynamics and performances, depending on the historical safe experience. The new policy consists of solving a constrained optimization problem, with two Gaussian Processes modelling, respectively, the safety constraints and the performance metric (i.e., objective function). As a probabilistic model, Gaussian Process regression provides an estimate of the target variable and the associated uncertainty; this property is crucial for dealing with uncertainty while new policies are safely explored. Another important benefit is that the proposed approach does not require any implementation of an expensive digital twin of the original system. Results on two real-life systems are presented, empirically proving the ability of the approach to improve performances with respect to the initial safe policy without significantly affecting safety.

Stochastic Programming is a powerful framework that addresses decision-making under uncertainties, which is a frequent occurrence in real-world problems. To effectively solve Stochastic Programming problems, scenario generation is one of the common practices that organizes realizations of stochastic processes with finite discrete distributions, which enables the use of mathematical programming models of the original problem. The quality of solutions is significantly influenced by the scenarios employed, necessitating a delicate balance between incorporating informative scenarios and preventing overfitting. Distributions-based scenario generation methodologies have been extensively studied over time, while a relatively recent concept of problem-driven scenario generation has emerged, aiming to incorporate the underlying problem’s structure during the scenario generation process. This survey explores recent literature on problem-driven scenario generation algorithms and methodologies. The investigation aims to identify circumstances under which this approach is effective and efficient. The work provides a comprehensive categorization of existing literature, supplemented by illustrative examples. Additionally, the survey examines potential applications and discusses avenues for its integration with machine learning technologies. By shedding light on the effectiveness of problem-driven scenario generation and its potential for synergistic integration with machine learning, this survey contributes to enhanced decision-making strategies in the context of uncertainties.

Modelling infectious disease spreading is crucial for planning effective containment measures, as shown in the COVID-19 pandemic. The effectiveness of planned measures can also be measured regarding saved lives and economic resources. Therefore, introducing methods able to model the evolution and the impact of measures, as well as planning tailored and updated measures, is a crucial step. Existing models for spreading modelling belong to two main classes: (i) compartmental models based on ordinary differential equations and (ii) contact-based models based on a contact structure using an underlining layer to simulate diffusion. Nevertheless, none of these methods can leverage the high computational power of artificial intelligence and deep learning. We propose a novel framework for simulating and analysing disease progression for these methods. The framework is based on the multiscale simulation of the spreading based on using a multiscale contact model built on top of a diffusion model customised by the user. The evolution of the spreading, modelled as a graph with attributed nodes, is then mapped into a latent space through graph embedding. Finally, deep learning models are used in the latent space to analyse and forecast methods without running expensive computational simulations of the contact-based model.

Climate change, desertification, salinisation of soils and the changing hydrology of the Earth are creating or modifying microbial habitats at all scales including the oceans, saline groundwaters and brine lakes. In environments that are saline or hypersaline, the biodegradation of recalcitrant plant and animal polysaccharides can be inhibited by salt-induced microbial stress and/or by limitation of the metabolic capabilities of halophilic microbes. We recently demonstrated that the chitinolytic haloarchaeon Halomicrobium can serve as the host for an ectosymbiont, nanohaloarchaeon ‘Candidatus Nanohalobium constans’. Here, we consider whether nanohaloarchaea can benefit from the haloarchaea-mediated degradation of xylan, a major hemicellulose component of wood. Using samples of natural evaporitic brines and anthropogenic solar salterns, we describe genome-inferred trophic relations in two extremely halophilic xylan-degrading three-member consortia. We succeeded in genome assembly and closure for all members of both xylan-degrading cultures and elucidated the respective food chains within these consortia. We provide evidence that ectosymbiontic nanohaloarchaea is an active ecophysiological component of extremely halophilic xylan-degrading communities (although by proxy) in hypersaline environments. In each consortium, nanohaloarchaea occur as ectosymbionts of Haloferax, which in turn act as scavenger of oligosaccharides produced by xylan-hydrolysing Halorhabdus. We further obtained and characterised the nanohaloarchaea–host associations using microscopy, multi-omics and cultivation approaches. The current study also doubled culturable nanohaloarchaeal symbionts and demonstrated that these enigmatic nano-sized archaea can be readily isolated in binary co-cultures using an appropriate enrichment strategy. We discuss the implications of xylan degradation by halophiles in biotechnology and for the United Nation's Sustainable Development Goals. © 2023 The Authors. Microbia

The use of electronic invoices helps to reduce the amount of paper used to store transaction information and offers many benefits to the companies. Indeed, it allows to speed the process, improve the efficiency, and reduce the environmental externalities. In addition, the availability of real-data on business operations ensures the timely detection of fraudulent transactions. However, the electronic invoicing process requires that all the invoices are converted to a standard electronic format, specified by local governments. In this work, we present a software tool developed to support the invoice dematerialization process in Italy. The implemented system is based on the named entity recognition model, a natural language processing technique capable of identifying the structure of the documents and automatically extrapolating the information. The developed tool has been tested by an Italian company. The results obtained are encouraging.

Driving is a complex activity which requires constant care and attention. Intelligent Advance Driver Assistance Systems (ADAS) can improve vehicle control performance and, thus, drivers and passengers safety. In particular, identification and prediction of driving intention can provide prompt information to drivers and vehicles in their vicinity that are fundamental for avoiding collisions. In this paper, we propose a lane change prediction model based on machine learning able to distinguish between left and right lane changes, a distinction that becomes particularly important when driving in a highway. Models have been trained and validated using a real dataset gathered online by using a high-tech demonstrator vehicle provided by Centro Ricerche Fiat (i.e., Fiat Research Center). Data, which refer to real driving conditions on a highway, have been collected by monitoring different drivers showing different behaviors. We address the problem of unbalanced data, typical of real data sets, and propose two prediction models based on Support Vector Machines and Random Forests. The results of our computational experiments show the validity of the approach with respect to state of the art models, both in terms of prediction accuracy and prediction time.

In this paper, we address the problem of automatic misogyny identification focusing on understanding the representation capabilities of widely adopted embeddings and addressing the problem of unintended bias. The proposed framework, grounded on Sentence Embeddings and Multi-Objective Bayesian Optimization, has been validated on an Italian dataset. We highlight capabilities and weaknesses related to the use of pre-trained language, as well as the contribution of Bayesian Optimization for mitigating the problem of biased predictions.

Telerehabilitation is a growing area of research and clinical practice which attempts to mitigate some of the major problems in chronic disease rehabilitation programs: short-staffed clinical care teams, great demand for complex face-to-face treatments, and lack of tools for reaching, monitoring and aiding target clinical populations. Telerehabilitation attempts to solve this through the use of easily accessible digital tools such as mobile and web-based applications which often rely on some form of data collection and analysis. Empirically tested perspectives on the integration of those data-driven tools in the real-time decision-making process of clinical care practitioners are still lacking. In this paper, we present a Decision Support System prototype, designed in the context of an applied game as a part of a comprehensive telerehabilitation software system, with the purpose of supporting real-time dynamic data visualization, understanding of patient gameplay and care routine patterns and, ultimately, enhancing the clinical care and design teams' decision- making processes.

In routine clinical practice, hepatitis C virus-infected patients can prematurely discontinue the prescribed regimen for several reasons. The aim of our study was to investigate sustained virological response (SVR12) rates in patients who prematurely discontinued directly acting antiviral (DAA) regimens and to assess the shortest effective duration of DAA able to lead to SVR12. We retrospectively collected the SVR rates of patients, registered in the NAVIGATORE-Lombardia Network database from January 2015, who discontinued DAAs before the predefined end of treatment. Overall, we included 365 patients, males were the majority (213, 58.4%), mean age was 60.5 years, and 53 (14.5%) patients were HIV-co-infected. Liver cirrhosis was observed in 251 (68.8%) subjects, and the most represented genotypes were 1b (n = 168, 46%) and 3 (n = 59, 16.2%). DAA was discontinued a median of 1 (IQR 1–4) weeks before the predefined EOT, with 164 (44.9%) patients stopping DAAs at least 2 weeks before the planned schedule. In patients with F0–F3 liver fibrosis, lower rates of SVR12 were observed in patients treated for <4 weeks: 50% (n = 2/4) vs. 99.1% (n = 109/110) for ≥4 weeks, p = 0.003. In patients with liver cirrhosis, lower rates of SVR12 were observed in patients treated <8 weeks: 83.3% (n = 25/30) vs. 94.6% (n = 209/221) for ≥8 weeks, p = 0.038. Despite premature discontinuation of DAA, high SVR12 rates were observed in a real-life setting for treatment lasting at least 4 weeks in patients with liver fibrosis F0–F3 and 8 weeks in those with liver cirrhosis. On this basis, feasibility of reducing DAA treatment duration should be explored in randomized clinical trials.

Despite the wide range of proposed biomarkers for Parkinson’s disease (PD), there are no specific molecules or signals able to early and uniquely identify the pathology onset, progression and stratification. Saliva is a complex biofluid, containing a wide range of biological molecules shared with blood and cerebrospinal fluid. By means of an optimized Raman spectroscopy procedure, the salivary Raman signature of PD can be characterized and used to create a classification model. Raman analysis was applied to collect the global signal from the saliva of 23 PD patients and related pathological and healthy controls. The acquired spectra were computed using machine and deep learning approaches. The Raman database was used to create a classification model able to discriminate each spectrum to the correct belonging group, with accuracy, specificity, and sensitivity of more than 97% for the single spectra attribution. Similarly, each patient was correctly assigned with discriminatory power of more than 90%. Moreover, the extracted data were significantly correlated with clinical data used nowadays for the PD diagnosis and monitoring. The preliminary data reported highlight the potentialities of the proposed methodology that, once validated in larger cohorts and with multi-centered studies, could represent an innovative minimally invasive and accurate procedure to determine the PD onset, progression and to monitor therapies and rehabilitation efficacy.

We describe a Smart Tourism System called SMARTCAL, designed in the context of a R&D project aimed at supporting the tourism development in Calabria (a region in the South of Italy). The system is designed by considering Points and Events of Interest (PEOI) and their relationship with the local transport systems and infrastructures. A proactive tourist tour planner algorithm is proposed to generate optimized itineraries based on static and dynamic profiling of the users. Social media data are also taken into account for recommendations.

Background: Sofosbuvir (SOF)-based regimens have been associated with renal function worsening in HCV patients with estimated glomerular filtration rate (eGFR) ≤ 45 ml/min, but further investigations are lacking. Aim: To assess renal safety in a large cohort of DAA-treated HCV patients with any chronic kidney disease (CKD). Methods: All HCV patients treated with DAA in Lombardy (December 2014–November 2017) with available kidney function tests during and off-treatment were included. Results: Among 3264 patients [65% males, 67% cirrhotics, eGFR 88 (9–264) ml/min], CKD stage was 3 in 9.5% and 4/5 in 0.7%. 79% and 73% patients received SOF and RBV, respectively. During DAA, eGFR declined in CKD-1 (p < 0.0001) and CKD-2 (p = 0.0002) patients, with corresponding rates of CKD stage reduction of 25% and 8%. Conversely, eGFR improved in lower CKD stages (p < 0.0001 in CKD-3a, p = 0.0007 in CKD-3b, p = 0.024 in CKD-4/5), with 33–45% rates of CKD improvement. Changes in eGFR and CKD distribution persisted at SVR. Baseline independent predictors of CKD worsening at EOT and SVR were age (p < 0.0001), higher baseline CKD stages (p < 0.0001) and AH (p = 0.010 and p < 0.0001, respectively). Conclusions: During DAA, eGFR significantly declined in patients with preserved renal function and improved in those with lower CKD stages, without reverting upon drug discontinuation.

Nowadays, the exponential diffusion of information forces Machine Learning algorithms to take relations into account in addition to data, which are no longer independent. We propose a Bayesian ensemble learning methodology named Relational Bayesian Model Averaging (RBMA) which, in addition to a probabilistic ensemble voting, takes relations into account. We tested the RBMA on a benchmark dataset for Sentiment Analysis in social networks and we compared it with its previous non-relational variant and we show that the introduction of relations significantly improves the performance of classification. Moreover, we propose a model for making predictions when new data becomes available modifying and increasing the underneath graph of relations on which the RBMA was trained.