Search

Images, texts, voices, and signals can be synthesized by latent spaces in a multidimensional vector, which can be explored without the hurdles of noise or other interfering factors. In this paper, we present a practical use case that demonstrates the power of latent space in exploring complex realities such as image space. We focus on DaVinciFace, an AI-based system that explores the StyleGAN2 space to create a high-quality portrait for anyone in the style of the Renaissance genius Leonardo da Vinci. The user enters one of their portraits and receives the corresponding Da Vinci-style portrait as an output. Since most of Da Vinci’s artworks depict young and beautiful women (e.g., “La Belle Ferroniere”, “Beatrice de’ Benci”), we investigate the ability of DaVinciFace to account for other social categorizations, including gender, race, and age. The experimental results evaluate the effectiveness of our methodology on 1158 portraits acting on the vector representations of the latent space to produce high-quality portraits that retain the facial features of the subject’s social categories, and conclude that sparser vectors have a greater effect on these features. To objectively evaluate and quantify our results, we solicited human feedback via a crowd-sourcing campaign. Analysis of the human feedback showed a high tolerance for the loss of important identity features in the resulting portraits when the Da Vinci style is more pronounced, with some exceptions, including Africanized individuals.

In a context of deep transformation of the entire automotive industry, starting from pervasive and native connectivity, commercial vehicles (heavy, light, and buses) are generating and transmitting much more data than passenger cars, with a much higher expected value, motivated by the higher costs of the vehicles and their added-value related businesses, such as logistics, freight, and transportation management. This paper presents a data-driven and unsupervised methodology to provide a descriptive model assessing the residual value estimates of heavy trucks subject to buy-back. A huge amount of telematics data characterizing the actual usage of commercial vehicles is jointly analyzed with different external conditions (e.g., altimetry), affecting the truck's performance to estimate the devaluation of the vehicle at the buy-back. The proposed approach has been evaluated on a large set of real-world heavy trucks to demonstrate its effectiveness in correctly assessing the real status of wear and residual value at the end of leasing contracts, to provide a few and quantitative insights through an informative, interactive and user-friendly dashboard to make a proper decision on the next business strategies to be adopted. The proposed solution has already been deployed by a private company within its data analytics services since (1) an interpretable descriptive model of the main factors/parameters and corresponding weights affecting the residual value is provided and (2) the experimental results confirmed the promising outcomes of the proposed data-driven methodology.

The overwhelming increase of parcel transports has prompted the need for effective and scalable intelligent logistics systems. In parallel, with the advent of Industry 4.0, a tight integration of Internet of Things technologies and Big Data analytics solution has become necessary to effectively manage industrial processes and to early predict product faults or service disruptions. In the context of good transports, the development of smart monitoring tools is particularly useful for couriers to ensure effective and efficient parcel deliveries. However, the existing predictive maintenance frameworks are not tailored to parcel delivery services. We present REDTag Service, an integrated framework to track and monitor the shipped packages. It relies on a network of IoT-enabled devices, called REDTags, allowing courier employees to easily collect the status of the package at each delivery step. The framework provides back-end functionalities for smart data transmission, management, storage, and analytics. A machine-learning process is included to promptly analyze the features describing event-related data to predict potential breaks of the goods in the packages. The framework provides also a dynamic view on the integrated data tailored to the different stakeholders, as well as on the prediction outcomes, enabling immediate feedback and model improvements. We analyze a real-world dataset including event-related data about parcel transports. To validate the hypothesis that the acquired data contains information relevant to predict the package status (i.e., broken or safe), we empirically analyze the performance of different, scalable classifiers. The experimental results confirm, in good approximation, the predictive power of the models extracted from the event-related features. To the best of the authors' knowledge, this work is the first attempt to address predictive maintenance in smart good transport logistics to predict package breaks from real-world data.

In recent years, the number and heterogeneity of large scientific datasets have been growing steadily. Moreover, the analysis of these data collections is not a trivial task. There are many algorithms capable of analyzing large datasets, but parameters need to be set for each of them. Moreover, larger datasets also mean greater complexity. All this leads to the need to develop innovative, scalable, and parameter-free solutions. The goal of this research activity is to design and develop an automated data analysis engine that effectively and efficiently analyzes large collections of text data with minimal user intervention. Both parameter-free algorithms and self-assessment strategies have been proposed to suggest algorithms and specific parameter values for each step that characterizes the analysis pipeline. The proposed solutions have been tailored to text corpora characterized by variable term distributions and different document lengths. In particular, a new engine called ESCAPE (enhanced self-tuning characterization of document collections after parameter evaluation) has been designed and developed. ESCAPE integrates two different solutions for document clustering and topic modeling: the joint approach and the probabilistic approach. Both methods include ad hoc self-optimization strategies to configure the specific algorithm parameters. Moreover, novel visualization techniques and quality metrics have been integrated to analyze the performances of both approaches and to help domain experts interpret the discovered knowledge. Both approaches are able to correctly identify meaningful partitions of a given document corpus by grouping them according to topics.

Wildfires are one of the natural hazards that the European Union is actively monitoring through the Copernicus EMS Earth observation program which continuously releases public information related to such catastrophic events. Such occurrences are the cause of both short- and long-term damages. Thus, to limit their impact and plan the restoration process, a rapid intervention by authorities is needed, which can be enhanced by the use of satellite imagery and automatic burned area delineation methodologies, accelerating the response and the decision-making processes. In this context, we analyze the burned area severity estimation problem by exploiting a state-of-the-art deep learning framework. Experimental results compare different model architectures and loss functions on a very large real-world Sentinel2 satellite dataset. Furthermore, a novel multi-channel attention-based analysis is presented to uncover the prediction behaviour and provide model interpretability. A perturbation mechanism is applied to an attention-based DS-UNet to evaluate the contribution of different domain-driven groups of channels to the severity estimation problem.

The pervasive and increasing deployment of smart meters allows collecting a huge amount of fine-grained energy data in different urban scenarios. The analysis of such data is challenging and opening up a variety of interesting and new research issues across energy and computer science research areas. The key role of computer scientists is providing energy researchers and practitioners with cutting-edge and scalable analytics engines to effectively support their daily research activities, hence fostering and leveraging data-driven approaches. This paper presents SPEC, a scalable and distributed engine to predict building-specific power consumption. SPEC addresses the full analytic stack and exploits a data stream approach over sliding time windows to train a prediction model tailored to each building. The model allows us to predict the upcoming power consumption at a time instant in the near future. SPEC integrates different machine learning approaches, specifically ridge regression, artificial neural networks, and random forest regression, to predict fine-grained values of power consumption, and a classification model, the random forest classifier, to forecast a coarse consumption level. SPEC exploits state-of-the-art distributed computing frameworks to address the big data challenges in harvesting energy data: the current implementation runs on Apache Spark, the most widespread high-performance data-processing platform, and can natively scale to huge datasets. As a case study, SPEC has been tested on real data of an heating distribution network and power consumption data collected in a major Italian city. Experimental results demonstrate the effectiveness of SPEC to forecast both fine-grained values and coarse levels of power consumption of buildings.

Energy performance certification is an important tool for the assessment and improvement of energy efficiency in buildings. In this context, estimating building energy demand also in a quick and reliable way, for different combinations of building features, is a key issue for architects and engineers who wish, for example, to benchmark the performance of a stock of buildings or optimise a refurbishment strategy. This paper proposes a methodology for (i) the automatic estimation of the building Primary Energy Demand for space heating ( P E D h ) and (ii) the characterization of the relationship between the P E D h value and the main building features reported by Energy Performance Certificates (EPCs). The proposed methodology relies on a two-layer approach and was developed on a database of almost 90,000 EPCs of flats in the Piedmont region of Italy. First, the classification layer estimates the segment of energy demand for a flat. Then, the regression layer estimates the P E D h value for the same flat. A different regression model is built for each segment of energy demand. Four different machine learning algorithms (Decision Tree, Support Vector Machine, Random Forest, Artificial Neural Network) are used and compared in both layers. Compared to the current state-of-the-art, this paper brings a contribution in the use of data mining techniques for the asset rating of building performance, introducing a novel approach based on the use of independent data-driven models. Such configuration makes the methodology flexible and adaptable to different EPCs datasets. Experimental results demonstrate that the proposed methodology can estimate the energy demand with reasonable errors, using a small set of building features. Moreover, the use of Decision Tree algorithm enables a concise interpretation of the quantitative rules used for the estimation of the energy demand. The methodology can be useful during both designing and refurbishment of buildings, to quickly estimate the expected building energy demand and set credible targets for improving performance.

Bicycle sharing systems are becoming increasingly popular in cities around the world as they are an inexpensive and sustainable means of transportation. Promoting the use of these systems substantially improves the quality of life in cities by reducing pollutant emissions and traffic congestion. In these systems, bikes are made available for shared use to individuals on a short-term basis. They allow people to borrow a bike in one dock and return it to any other station with free docks belonging to the same system. The occupancy level of the stations can be constantly monitored. However, to achieve a satisfactory user experience, all the stations in the system must be neither overloaded nor empty when the user needs to access the station. The aim of this paper is to analyze occupancy level data acquired from real systems to determine situations of dock overload in multiple stations which could lead to service disruption. The proposed methodology relies on a pattern mining approach. A new pattern type called Occupancy Monitoring Pattern is proposed here to detect situations of dock overload in multiple stations. Since stations are geo-referenced and their occupancy levels are periodically monitored, occupancy patterns can be filtered and evaluated by taking into consideration both the spatial and temporal correlation of the acquired measurements. The results achieved on real data highlight the potential of the proposed methodology in supporting domain experts in their maintenance activities, such as periodic re-balancing of the occupancy levels of the stations, as well as in improving user experience by suggesting alternative stations in the nearby area.

In the last few years, a large number of smart meters have been deployed in buildings to continuously monitor fine-grained energy consumption. Meteorological data deeply impact energy consumption, and an in-depth analysis of collected and correlated data can uncover interesting and actionable insights to improve the overall energy balance of our communities and to enhance people’s awareness of energy wasting. To effectively extract meaningful and interpretable insights from large collections of energy measurements and multi-dimensional meteorological data, innovative data science methodologies should be devised. Research frontiers are addressing self-learning approaches, which allow non-experts to exploit machine learning techniques more easily, and algorithmic transparency of models, hence providing actionable, explicit, declarative knowledge representation. This paper presents METeorological Data Analysis for Thermal Energy CHaracterization (METATECH), a data mining engine based on both exploratory and unsupervised data analytics algorithms, devised to build transparent models correlating weather conditions and energy consumption in buildings. METATECH exploits a joint approach coupling cluster analysis and generalized association rules to allow a deeper yet human-readable understanding of how meteorological data impact heating consumption. First, a partitional clustering algorithm is applied to weather conditions. Then, resulting clusters are characterized by means of generalized association rules, which provide a self-learning explainable model of the most interesting correlations between energy consumption and weather conditions at different granularity levels. The experimental evaluation performed on real datasets demonstrates the effectiveness of the proposed approach in automatically extracting interesting knowledge from data, and provide it transparently to domain experts.