Search

IntroductionArtificial Intelligence (AI) methods are being increasingly investigated as a means to generate predictive models applicable in the clinical practice. In this study, we developed a model to predict the efficacy of immunotherapy (IO) in patients with advanced non-small cell lung cancer (NSCLC) using eXplainable AI (XAI) Machine Learning (ML) methods.MethodsWe prospectively collected real-world data from patients with an advanced NSCLC condition receiving immune-checkpoint inhibitors (ICIs) either as a single agent or in combination with chemotherapy. With regards to six different outcomes - Disease Control Rate (DCR), Objective Response Rate (ORR), 6 and 24-month Overall Survival (OS6 and OS24), 3-months Progression-Free Survival (PFS3) and Time to Treatment Failure (TTF3) - we evaluated five different classification ML models: CatBoost (CB), Logistic Regression (LR), Neural Network (NN), Random Forest (RF) and Support Vector Machine (SVM). We used the Shapley Additive Explanation (SHAP) values to explain model predictions.ResultsOf 480 patients included in the study 407 received immunotherapy and 73 chemo- and immunotherapy. From all the ML models, CB performed the best for OS6 and TTF3, (accuracy 0.83 and 0.81, respectively). CB and LR reached accuracy of 0.75 and 0.73 for the outcome DCR. SHAP for CB demonstrated that the feature that strongly influences models’ prediction for all three outcomes was Neutrophil to Lymphocyte Ratio (NLR). Performance Status (ECOG-PS) was an important feature for the outcomes OS6 and TTF3, while PD-L1, Line of IO and chemo-immunotherapy appeared to be more important in predicting DCR.ConclusionsIn this study we developed a ML algorithm based on real-world data, explained by SHAP techniques, and able to accurately predict the efficacy of immunotherapy in sets of NSCLC patients.

Keeping students engaged while learning programming is becoming more and more imperative. Of the several proposed techniques, gamification is presumably the most widely studied and has already proven as an effective means to engage students. However, there is a complete lack of public and customizable solutions to gamified programming education that can be reused with personalized rules and learning material. FGPE Gamification Service (FGPE GS) is an open-source GraphQL service that transforms a package containing the gamification layer – adhering to a dedicated open-source language, GEdIL – into a game. The game provides students with a gamified experience leveraging on the automatically-assessable activities referenced by the challenges. This paper presents FGPE GS, its architecture, data model, and validation.

An increasing number of organisations are harnessing the benefits of hybrid cloud adoption to support their business goals and achieving privacy and control in a private cloud whilst enjoying the on-demand scalability of the public cloud. However the complexity introduced by the combination of the public and private clouds worsens visibility in cloud monitoring with regards compliance to given business constraints. Load balancing as a technique for evenly distributing workloads can be leveraged together with processing mining to help ease the monitoring challenge. In this paper we propose a load balancing approach to distribute workloads in order to minimise violations to specified business constraints. The scenario of a hospital consultation process is employed as a use case in monitoring and controlling Octavia load balancing-as-a-service in OpenStack. The results show a co-occurrence of constraint violations and Octavia L7 Policy creation, indicating a successful application of process mining monitoring in load balancing.

Conventional recommender systems provide personalized recommendations by collecting and retaining user data, relying on a centralized architecture. Hence, user privacy is undermined by the volume of information required to support the personalized experience. In this work, we propose a User Experience model which allows the privacy of a user to be preserved by means of a decentralized architecture, enabling the Service Provider to offer recommendations without the need of storing individual user data. We advance the current state of the art by: i) Proposing a model of User Experience (UEx) suitable for Persona-based recommendations; ii) Presenting a UEx collection model which enhances the user privacy towards the service provider while keeping the quality of her preferences predictions; and iii) Assessing the existence of the Persona profiles, which are needed for generating and addressing the recommendations. We perform several experiments using a real-world complete dataset from a medium-sized service provider, composed of more than 14,000 unique users and 33,000 content titles collected over a period of two years. We show that our architecture, in combination with our UEx model, achieves the same or better results, compared to state-of-the-art systems, in terms of rating prediction accuracy, without sacrificing user’s privacy.

Trading cryptocurrency on current digital exchange platforms is a trust-based process, where the parties involved in the exchange have to fully trust the service provider. As it has been proven several times, this could lead to funds being stolen, either due to malicious service providers that simply disappear or due to hacks that these platforms might suffer. In this work, we propose and develop a decentralised exchange solution based on smart contracts running on the Ethereum network that is open, verifiable, and does not require trust. The platform enables two parties to trade different currencies, limited to Ethereum and Bitcoin in the current status of the system. A smart contract, deployed on the Ethereum blockchain, functions as an escrow, which holds a user’s funds until a verified transaction has been made by the other party. To make the smart contract able to detect a Bitcoin transfer, we implement our solution by utilising an oracle. We define the system architecture and implement a working platform, which we test in a model scenario, successfully exchanging Bitcoin and Ether on the blockchain test networks. We conclude the paper identifying possible challenges and threats to such a system.

Low-power devices are usually highly constrained in terms of CPU computing power, memory, and GPGPU resources for real-time applications to run. In this paper, we describe RAPID, a complete framework suite for computation offloading to help low-powered devices overcome these limitations. RAPID supports CPU and GPGPU computation offloading on Linux and Android devices. Moreover, the framework implements lightweight secure data transmission of the offloading operations. We present the architecture of the framework, showing the integration of the CPU and GPGPU offloading modules. We show by extensive experiments that the overhead introduced by the security layer is negligible. We present the first benchmark results showing that Java/Android GPGPU code offloading is possible. Finally, we show the adoption of the GPGPU offloading into BioSurveillance , a commercial real-time face recognition application. The results show that, thanks to RAPID, BioSurveillance is being successfully adapted to run on low-power devices. The proposed framework is highly modular and exposes a rich application programming interface to developers, making it highly versatile while hiding the complexity of the underlying networking layer.

It was just a few years ago when I bought my first smartphone. And now, (almost) all of my friends possess at least one of these powerful devices. International Data Corporation (IDC) reports that smartphone sales showed strong growth worldwide in 2011, with 491.4 million units sold – up to 61.3 percent from 2010. Furthermore, IDC predicts that 686 million smartphones will be sold in 2012, 38.4 percent of all handsets shipped. Silently, we are becoming part of a big mobile smartphone network, and it is amazing how the perception of the world is changing thanks to these small devices. If many years ago the birth of Internet enabled the possibility to be online, smartphones nowadays allow to be online all the time. Today we use smartphones to do many of the tasks we used to do on desktops, and many new ones. We browse the Internet, watch videos, upload data on social networks, use online banking, find our way by using GPS and online maps, and communicate in revolutionary ways. Along with these benefits, these fancy and exciting devices brought many challenges to the research area of mobile and distributed systems. One of the first problems that captured our attention was the study of the network that potentially could be created by interconnecting all the smartphones together. Typically, these devices are able to communicate with each other in short distances by using com- munication technologies such as Bluetooth or WiFi. The network paradigm that rises from this intermittent communication, also known as Pocket Switched Network (PSN) or Opportunistic Network ([10, 11]), is seen as a key technology to provide innovative services to the users without the need of any fixed infrastructure. In PSNs nodes are short range communicating devices carried by humans. Wireless communication links are created and dropped in time, depending on the physical distance of the device holders. From one side, social relations among humans yield recurrent movement patterns that help researchers design and build protocols that efficiently deliver messages to destinations ([12, 13, 14] among others). The complexity of these social relations, from the other side, makes it difficult to build simple mobility models, that in an efficient way, generate large synthetic mobility traces that look real. Traces that would be very valuable in protocol validation and that would replace the limited experimentally gathered data available so far. Traces that would serve as a common benchmark to researchers worldwide on which to validate existing and yet to be designed protocols. With this in mind we start our study with re-designing SWIM [15], an already exist- ing mobility model shown to generate traces with similar properties of that of existing real ones. We make SWIM able to easily generate large (small)-scale scenarios, starting from known small (large)-scale ones. To the best of our knowledge, this is the first such study. In addition, we study the social aspects of SWIM-generated traces. We show how to SWIM-generate a scenario in which a specific community structure of nodes is required. Finally, exploiting the scaling properties of SWIM, we present the first analysis of the scal- ing capabilities of several forwarding protocols such as Epidemic [16], Delegation [13], Spray&Wait [14], and BUBBLE [12]. The first results of these works appeared in [1], and, at the time of writing, [2] is accepted with minor revision. Next, we take into account the fact that in PSNs cannot be assumed full cooperation and fairness among nodes. Selfish behavior of individuals has to be considered, since it is an inherent aspect of humans, the device holders (see [17], [18]). We design a market-based mathematical framework that enables heterogeneous mobile users in an opportunistic mobile network to compromise optimally and efficiently on their QoS 3 demands. The goal of the framework is to satisfy each user with its achieved (lesser) QoS, and at the same time maximize the social welfare of users in the network. We base our study on the consideration that, in practice, users are generally tolerant on accepting lesser QoS guarantees than what they demand, with the degree of tolerance varying from user to user. This study is described in details in Chapter 2 of this dissertation, and is included in [3]. In general, QoS could be parameters such as response time, number of computations per unit time, allocated bandwidth, etc. Along the way toward our study of the smartphone-world, there was the big advent of mobile cloud computing—smartphones getting help from cloud-enabled services. Many researchers started believing that the cloud could help solving a crucial problem regarding smartphones: improve battery life. New generation apps are becoming very complex — gaming, navigation, video editing, augmented reality, speech recognition, etc., — which require considerable amount of power and energy, and as a result, smartphones suffer short battery lifetime. Unfortunately, as a consequence, mobile users have to continually upgrade their hardware to keep pace with increasing performance requirements but still experience battery problems. Many recent works have focused on building frameworks that enable mobile computation offloading to software clones of smartphones on the cloud (see [19, 20] among others), as well as to backup systems for data and applications stored in our devices [21, 22, 23]. However, none of these address dynamic and scalability features of execution on the cloud. These are very important problems, since users may request different computational power or backup space based on their workload and deadline for tasks. Considering this and advancing on previous works, we design, build, and implement the ThinkAir framework, which focuses on the elasticity and scalability of the server side and enhances the power of mobile cloud computing by parallelizing method execution using multiple Virtual Machine (VM) images. We evaluate the system using a range of benchmarks starting from simple micro-benchmarks to more complex applications. First, we show that the execution time and energy consumption decrease two orders of magnitude for the N-queens puzzle and one order of magnitude for a face detection and a virus scan application, using cloud offloading. We then show that a parallelizable application can invoke multiple VMs to execute in the cloud in a seamless and on-demand manner such as to achieve greater reduction on execution time and energy consumption. Finally, we use a memory-hungry image combiner tool to demonstrate that applications can dynamically request VMs with more computational power in order to meet their computational requirements. The details of the ThinkAir framework and its evaluation are described in Chapter 4, and are included in [6, 5]. Later on, we push the smartphone-cloud paradigm to a further level: We develop Clone2Clone (C2C), a distributed platform for cloud clones of smartphones. Along the way toward C2C, we study the performance of device-clones hosted in various virtualization environments in both private (local servers) and public (Amazon EC2) clouds. We build the first Amazon Customized Image (AMI) for Android-OS—a key tool to get reliable performance measures of mobile cloud systems—and show how it boosts up performance of Android images on the Amazon cloud service. We then design, build, and implement Clone2Clone, which associates a software clone on the cloud to every smartphone and in- terconnects the clones in a p2p fashion exploiting the networking service within the cloud. On top of C2C we build CloneDoc, a secure real-time collaboration system for smartphone users. We measure the performance of CloneDoc on a testbed of 16 Android smartphones and clones hosted on both private and public cloud services and show that C2C makes it possible to implement distributed execution of advanced p2p services in a network of mobile smartphones. The design and implementation of the Clone2Clone platform is included in [7], recently submitted to an international conference. We believe that Clone2Clone not only enables the execution of p2p applications in a network of smartphones, but it can also serve as a tool to solve critical security problems. In particular, we consider the problem of computing an efficient patching strategy to stop worm spreading between smartphones. We assume that the worm infects the devices and spreads by using bluetooth connections, emails, or any other form of communication used by the smartphones. The C2C network is used to compute the best strategy to patch the smartphones in such a way that the number of devices to patch is low (to reduce the load on the cellular infrastructure) and that the worm is stopped quickly. We consider two well defined worms, one spreading between the devices and one attacking the cloud before moving to the real smartphones. We describe CloudShield [8], a suite of protocols running on the peer-to-peer network of clones; and show by experiments with two different datasets (Facebook and LiveJournal) that CloudShield outperforms state-of-the-art worm-containment mechanisms for mobile wireless networks. This work is done in collaboration with Marco Valerio Barbera, PhD colleague in the same department, who contributed mainly in the implementation and testing of the malware spreading and patching strategies on the different datasets. The communication between the real devices and the cloud, necessary for mobile com- putation offloading and smartphone data backup, does certainly not come for free. To the best of our knowledge, none of the works related to mobile cloud computing explicitly studies the actual overhead in terms of bandwidth and energy to achieve full backup of both data/applications of a smartphone, as well as to keep, on the cloud, up-to-date clones of smartphones for mobile computation offload purposes. In the last work during my PhD—again, in collaboration with Marco Valerio Barbera—we studied the feasibility of both mobile computation offloading and mobile software/data backup in real-life scenarios. This joint work resulted in a recent publication [9] but is not included in this thesis. As in C2C, we assume an architecture where each real device is associated to a software clone on the cloud. We define two types of clones: The off-clone, whose purpose is to support computation offloading, and the back-clone, which comes to use when a restore of user’s data and apps is needed. We measure the bandwidth and energy consumption incurred in the real device as a result of the synchronization with the off-clone or the back-clone. The evaluation is performed through an experiment with 11 Android smartphones and an equal number of clones running on Amazon EC2. We study the data communication overhead that is necessary to achieve different levels of synchronization (once every 5min, 30min, 1h, etc.) between devices and clones in both the off-clone and back-clone case, and report on the costs in terms of energy incurred by each of these synchronization frequencies as well as by the respective communication overhead. My contribution in this work is focused mainly on the experimental setup, deployment, and data collection.

Smartphones have exploded in popularity in recent years, becoming ever more sophisticated and capable. As a result, developers worldwide are building increasingly complex applications that require ever increasing amounts of computational power and energy. In this paper we propose ThinkAir, a framework that makes it simple for developers to migrate their smartphone applications to the cloud. ThinkAir exploits the concept of smartphone virtualization in the cloud and provides method level computation offloading. Advancing on previous works, it focuses on the elasticity and scalability of the server side and enhances the power of mobile cloud computing by parallelizing method execution using multiple Virtual Machine (VM) images. We evaluate the system using a range of benchmarks starting from simple micro-benchmarks to more complex applications. First, we show that the execution time and energy consumption decrease two orders of magnitude for the N-queens puzzle and one order of magnitude for a face detection and a virus scan application, using cloud offloading. We then show that if a task is parallelizable, the user can request more than one VM to execute it, and these VMs will be provided dynamically. In fact, by exploiting parallelization, we achieve a greater reduction on the execution time and energy consumption for the previous applications. Finally, we use a memory-hungry image combiner tool to demonstrate that applications can dynamically request VMs with more computational power in order to meet their computational requirements., 17 pages

Summary Fast technology development has influenced the widespread use of low-power devices in different scientific, environmental, and everyday life areas, giving birth to the Internet of Things. In this paper, we focus on the context of marine studies, addressing the problem of marine bathymetry data processing and analysis via pervasive and Internet-connected sensors and low-power distributed devices. Pervasive and Internet-connected low-power devices (as the components involved in the sensing and processing actions) made diverse and different ?things? as a worldwide-distributed system. Given the high complexity of the algorithms involved in these studies, which usually involve general-purpose graphic processing unit (GPGPU) computation, it is impossible for the limited devices to perform the required calculations. To overcome these limitations, in this paper, we propose and implement a vertical application of GVirtuS, the open-source GPGPU virtualization and remoting service, for achieving high performance geographical data interpolation in a high performance cloud computing scenario. We present an innovative implementation by comparing, in terms of performance and accuracy, the inverse distance weighting and kriging interpolation methods in their parallel implementations leveraging on CUDA-enabled GPGPUs. We present a real-world use case related to high-resolution bathymetry interpolation in a crowdsource data context in the Bay of Pozzuoli, Italy.