Schedule

Abstract

In a structured software development process, requirements are the the primary landmark of any development activity, from preliminary design to final system validation. Along the process, different stakeholders with diverse skills are supposed to cope with the requirements, and since natural language (NL) is the most largely understood communication code, requirements are typically expressed as NL documents. Given the inherent ambiguity of NL, techniques have been developed to minimise the ambiguity of the requirements. The course will present the typical ambiguity problems of NL requirements, and will introduce to the state-of-the-art techniques to support NL requirements analysis. In particular, part of the course will be dedicated to the presentation of tools for natural language processing -- i.e., Python NLTK and GATE -- which can be used to support ambiguity detection in NL requirements.

Schedule

Abstract

The course aims to provide useful guidelines and advices for writing a scientific paper.

Schedule

Abstract

The course deals with optimization methods underlying machine learning models. Supervised, unsupervised and semi-supervised learning models will be considered. Decomposition methods and incremental algorithms for Support Vector Machines and Neural Networks will be analyzed. Feature selection and feature ranking techniques based on nonlinear optimization algorithms will be presented. Optimization methods in big data will be analyzed and discussed.

Schedule

Abstract

The ever-increasing demand for reliable and ubiquitous high-speed data communications and environment sensing services calls for new challenges in the design and the optimization of wireless networks, which may benefit from the adoption of sophisticated signal processing techniques at large. Recently, game theory has emerged as an effective framework for the network design, since it provides analytical tools to predict the outcome of interactions among rational entities. This tutorial provides an overview of the relevant applications of game theory, focusing on state-of-the-art techniques for resource allocation in wireless and wired communication networks. In the first part, the very basics concepts are introduced by means of many simple examples, and special emphasis is put on how to translate a real-world problem to an analytical game model. In the second part, relevant applications of game theory to wireless networks design are reported, including power and rate control, bandwidth allocation, and spectrum sensing. Some clues will be given on how to extend such methods to MIMO, cognitive radio, and relay-assisted communications. The main focus will be on noncooperative techniques, although recent advances in the field of cooperative game theory will be also included in the discussion to provide a different perspective on certain classes of problems.

Schedule

Abstract

This short course will touch some basic and advanced topics in deep learning including: representations and feature learning, feedforward networks, supervised learning, optimization: structure of the problem, basic and advanced algorithms, tricks of the trade, regularization, convolutional networks, image recognition, adversarial training, autoencoders, energy-based models, recurrent networks.

Schedule

Abstract

TBA

Schedule

Abstract

In this short course we will expose methods used in reliability, availability, performability and survivability modeling and analysis of systems in practice. Non-state-space solution methods are often used to solve reliability block diagrams, fault trees and reliability graphs. Relatively efficient algorithms are known to handle systems with hundreds of components and have been implemented in many software packages. We will show the usage of these model types through practical examples and via the software package SHARPE. Nevertheless many practical problems cannot be handled by such algorithms. Bounding algorithms are then used in such cases as was done for a major subsystem of Boeing 787. Non-state-space methods derive their efficiency from the independence assumption that is often violated in practice. State space methods based on Markov chains, stochastic Petri nets, semi-Markov and Markov regenerative processes can be used to capture various kinds of dependencies among system components. Markov models, Markov Reward models and stochastic Petri nets will be illustrated through practical problems and using the SHARPE software package. However, the resulting state space explosion severely restricts the size of the problems that can be solved. Hierarchical and fixed-point iterative methods provide a scalable alternative that combines the strengths of state space and non-state-space methods and have been extensively used to solve real-life problems. The use of hierarchical and fixed point iterative methods will be also illustrated via large system examples and the SHARPE software package.

Schedule

Abstract

The aim of this course is to provide Computer Engineers with the theory and practical insight to obtain meaningful and sound performance result using simulation as an investigation tool. The course will introduce the coding practices to write down a discrete-event simulator. Then, it will describe the methodology to be followed when a large-scale simulation project is undertaken, including how to gather output data. The practical part of the course will consist in getting to know the Omnet++ simulation framework and its modular, message-based modeling paradigm.

Schedule

Abstract

Big Data analytics is gaining large interest in both public and scientific agendas, because it has demonstrated that it is possible to extract valid information from a large amount of noisy data and to produce valuable information for decision makers. Applications of Big Data analytics can be found in a large variety of domains, including economics, physics, healthcare and biology. In this last domain, analytics have been used, for example, to predict climate change impact on species' distribution, to monitor the effect of overfishing on economy and marine biodiversity and to prevent ecosystems collapse. In this course, practical applications of Big Data analytics will be shown, with focus on several signal processing and machine learning-based techniques. The course will clarify the general concepts behind these techniques, with an educational approach making these concepts accessible also to students with intermediate mathematical skills. The examples will regard real cases involving data that would have been unpractical to be human-analyzed and corrected, especially in the biology domain: time series forecasting, periodicities detection, comparison of geographical distribution maps, assessment of environmental similarities between different areas, global scale species distributions. The above techniques have a general purpose applicability and the students will be able to use them in other domains too. Cloud computing, data sharing, experiments reproducibility, usage of data representation standards and most of the requirements of Big Data analytics systems will be explained and practiced. To execute the experiments, students will use a distributed e-Infrastructure (D4Science) developed at ISTI-CNR, also used in the European Laboratory on Big Data Analytics and Social Mining (SoBigData). This web-based platform hides the complexity of implementing Big Data analytics processes from scratch and allows students to concentrate on experiments configuration and output evaluation, and to understand models' behaviours. For this reason, the course does not require any programming skill and is suited for students in Computer Engineering, Informatics, Telecommunications engineering, Statistics and Computational Biology.

Schedule

Abstract

The course will focus on recent advances in electronics and ICT (Information Communication Technology) for Autonomous Driver Assistance Systems in new vehicle generations. Sensing technologies and HW-SW embedded systems (architecture, including mm-wave transceivers and antennas, and algorithm) for parking assistance, enhanced driver vision, collision avoidance, cruise control, tyre condition monitoring, will be discussed. To this aim, recent state of art in video- and radar-based systems will be reviewed. Automotive and railway example applications will be provided.

Schedule

Abstract

TBA

Schedule

Abstract

This short course will cover a basic introduction to theory and algorithms for global optimization. It will be divided into the following sub-topics. 1) Basic definitions; problems which are NOT global optimization ones (a quick overview of convex optimization). 2) Classical and recent heuristic algorithms for large-scale global optimization. 3) Exact methods - Branch & Bound approaches for global optimization. 4) Heuristic methods for black box, expensive functions.

Schedule

Abstract

When observing a visual content on a web site, often people do not realize that such media have undergone a long series of transformations before appearing in their current form. In particular, the authenticity of visual contents is often overestimated: just by looking around, it is evident that creating fake digital contents is very common today, since many commercial and open-source software allow to inexpert users editing images and videos in a few minutes. Multimedia forensics is a relatively new discipline that seeks to demonstrate the authenticity of a given multimedia content: the basic idea behind it is that any processing applied to a digital content leaves subtle traces, that can be analyzed to uncover the digital history of the object. In the first part of the course, a general introduction to the field of multimedia forensics will be given. Then, the most useful traces that can be used for image tampering detection, and the main forensic techniques that are based on them, will be described.

Schedule

Abstract

Recently, researchers came up with a new security approach, from information theory fundamentals, which is focused on the exploitation of the intrinsic property of the radio environment and of the electromagnetic waves to produce security at physical layer, with no assumption on the computational capabilities of eavesdroppers and with the added benefit of reducing the load on the security protocols at the upper levels.  The course will give the fundamentals of physical layer security approach for wireless networks, and in particular, how it can help not only by adding security at one layer, but also in other aspects like key distribution.

Schedule

Abstract

The course will present the finite element method (FEM) starting from its storical development and general impact in engineering. It will then present the classes of problems which can be solved with it, presenting several examples, spanning from electric machines examples, to high frequency electromagnetics and thermal or civil engineering. Part of the course is devoted to advanced techniques for speeding up computation.

Schedule

Abstract

In this course, we first present facts and figures that highlight the importance of green mobile networking and then review existing green cellular networking research, providing general guidelines and design criteria in order to reduce power consumption, to increase the energy efficiency in cellular networks and to generalize these strategies to ICT systems. A particular focus will be on techniques that incorporate the concept of the “sleep mode” in base stations. It takes advantage of changing traffic patterns on daily or weekly basis and selectively switches some lightly loaded base stations to low energy consumption modes. As base stations are responsible for the large amount of energy consumed in cellular networks, these approaches have the potential to save a significant amount of energy.

Schedule

Abstract

"The presentation will touch upon the entire writing process of an academic publication and addresses the questions that arise in the mind of students who are new to the realm of publishing. The session is aimed at graduate and PHD students, who are working on research projects and seek recognition of their research through IEEE publications. The seminar will provide an insight into the following topics: • Overview of the structure of a scientific article and methods foi improving the writing skills, • Choosing where to communicate – a conference or a journal? How to decide? • A review of the editorial process will follow – “do’s” and “don’ts” during the submission and revision process. • Possibilities of open access publishing with IEEE • Open discussion on issues faced during the process, suggestions and sharing"

Schedule

Abstract

Schedule

Abstract

Automatic image annotation is an important task, in which the goal is to determine the relevance of annotation terms for images. To this end, a common and successful approach is to quantize local visual features (e.g. SIFT) following the well-known bag-of-visual-words paradigm. Then, a classifier (e.g. SVM) can be learned from a collection of images manually labeled as belonging to an object category or not. The most recent breakthrough in computer vision came from end-to-end feature learning through convolutional neural networks (CNN). These deep learning based architectures allow to learn image representations and classifiers jointly directly from pixels, without requiring any feature engineering. The participants will be guided to implement an image recognition system in Matlab, based on both handcrafted bag-of-visual-words and end-to-end learned representations. The emphasis of the course will be on the important general concepts rather than in depth coverage of contemporary papers.

Schedule

Abstract

"How to ensure that critical infrastructures, like power plants, transportation systems, data centers, and medical devices work safely and reliably? That is the topic of risk management, and fault tree analysis (FTA) is a very prominent technique here. FTA comprises a variety of methods and is very popular in practice, being deployed by many companies, like NASA, ESA, Honeywell, Airbus, Toyota, etc. This tutorial addresses how fault trees can be used to model and analyze dependability aspects of complex systems: I will give an introduction to the most common quantitative and qualitative analysis techniques, their practical relevance and deployment in practice. Moreover, I will illustrate how formal methods help to increase the modeling and analytical power of FTA. I will end by discussing fault tree extensions, as well as current research directions in FTA."

Schedule

Abstract

Schedule

Abstract

Schedule

Abstract

Software Product Line Engineering (SPLE) is a software engineering methodology aimed at developing, in a cost-effective and time-efficient manner, a family of software-intensive (highly configurable) systems by systematic reuse. Individual products (or variants) share an overall reference model or architecture of the family, but they differ with respect to specific features, which are (user-visible) increments in functionality. The explicit introduction and management of feature-based variability in the software development cycle causes complexity in the modelling and analysis of SPLs, because the number of possible individual products may be exponential in the number of features. Recent research has focussed on lifting successful formal specification languages and formal verification techniques known from single (software) system engineering, such as process calculi and model checking, to (software) product lines or highly configurable systems in general. In this course, after recalling several modelling and analysis techniques and a general introduction to SPLE, we present the formal underpinnings of a number of modelling and analysis frameworks that were introduced for the specification and verification of variability in behavioural models of SPLs. These range from variants of Modal Transition Systems (MTSs) and Featured Transitions Systems (FTSs) with (action-based, feature-based) CTL logics and their dedicated family-based model checkers VMC and SNIP to a feature mu-calculus over FTSs that allows family-based model checking with off-the-shelf model checkers like mCRL2.

Schedule

Abstract

Nonlinear manifolds have a special place in problem solutions where constraints of the problems restrict the domains to some interesting, structured sets. The differential geometry of these constrained spaces, or manifolds, guides us to reach more efficient solutions. Besides being mathematically appealing, the solutions based on the geometry of the underlying manifolds are often faster and more stable than their constrained optimization counterparts. This fact has been exploited in many branches of science and engineering, in developing methodologies, algorithms, and systems. In this short course, we will focus on several manifolds including shape manifolds of planar closed curves, Grassmann manifolds, and manifolds of covariance matrices (Symmetric Positive Semi Definite) and affine transformations. In each case, we will provide a mathematical background and demonstrate the use of these manifolds in 3D Computer Vision applications such as shape analysis, human behavior understanding, and activity classification from 3D data. Practical examples will be presented using Matlab.

Schedule

Abstract

(1) Introduction to propositional and first order logic. (2) fuzzy logics (3) variational Markov logic (4) Logic Tensor Networks (4) Markov logic networks and variational Markov logic.

Schedule

Abstract

Introduction. What’s new in IoT cybersecurity? Non-problems, old problems, new problems. Cybersecurity requirements, threats and vulnerabilities. Botnet of things. Crash course in applied cryptography (if necessary) | advanced concepts in applied cryptography. Security Issues in Typical IoT Protocols. IEEE 802.15.4. TISCH mode. 6LowPan and RPL. DTLS and CoAP Secure Localization in IoT. Distance bounding. Threats and vulnerabilities. Secure multilateration. Using drones and swarms of drones for secure multilateration. Evaluation. Presentation and discussion of a paper.

Schedule

Abstract

This short course aims to provide both theoretical and practical tools to tackle estimation problems encountered in several areas of engineering and science. In particular, it is shown how to formulate such estimation problems as instances of a general dynamical system state estimation problem and how to derive the mathematical solution of the latter problem. Then it is shown that, for a linear Gaussian system, such a solution yields the well known Kalman filter. Further, approximate techniques (e.g. extended and unscented Kalman filters, particle filter, etc.) are presented for the case of nonlinear and/or non-Gaussian systems, for which an exact closed-form solution cannot be found. To conclude the theoretical part, theoretical limitations (i.e. the Cramer-Rao lower bound) on the quality of estimation are discussed. In the second part of the course, we illustrate some applications of linear/nonlinear Kalman filtering (e.g., tracking, robotic navigation, environmental data assimilation).

Schedule

Abstract

"GPU architecture. Comparison w.r.t CPU architecture. SIMT. What is CUDA. Hello World in CUDA. CUDA Memory organization: introduction. Threads organization and setup. Exercises. Grids, blocks and threads. Thread scheduling. Threads and divergence. Good practices in blocks and threads organization. 2D CUDA kernels. 3D CUDA kernels. CUDA memory hierarchy. Shared, device and constant memory. Efficient CUDA programming. Tiling: introduction. Tiling: how to. Synchronization barriersd. Phase-based processing. Border checking. Performance considerations. DRAM burst. Memory coalescing. Atomic instructions. Privatization. Exercises: Histogram computation. 1D convolution. How to use constant memory.. Tiling 2D convolution. Tiling. Tiling performance evaluation. CUDA parallel patterns: Convolution/Stencil. Reduction. Scan. Lab activity: 2D convolution for image filtering. Handling 2D borders. Using constant memory for convolution weights. Conditions for tiling and computation."

Schedule

Abstract

The aim of Document Image Analysis is to extract information from digitized or born-digital documents. Originally conceived for recognizing text in printed documents these techniques initially contributed to the design of Optical Character Recognition systems. Algorithms and techniques developed in this area have been often extended to other Pattern Recognition application areas. Nowadays the research addresses several different areas like handwritten historical documents, scene text recognition and information extraction from born-digital documents among the others. In the first part of this course we will address traditional techniques for extracting information from document images discussing the main approaches for performing pre-processing, layout analysis and the recognition of characters and words from digitized and digital-born documents. In the second part we will cover recent applications and techniques in the field.

Schedule

Abstract

The development of software in safety-critical domains is subject to specific guidelines that aim at reducing the possibility that bugs in the software threat safety of the users. The course will introduce the software safety guidelines adopted in three transportation domains, namely railway signailling, avionics, automotive, and will discuss the differences among them and their actual implications on the software development process.

Schedule

Abstract

"Neural networks foundations: MLP, creating a simple network with Keras, towards deep learning Keras installation and API Deep learning with ConvNets, DCNN layers, experimenting with basic datasets, very deep networks Generative Adversarial Networks, GANs and images."

Schedule

Abstract

The course provides a brief overview and examples of resource allocation policies in wireless networks. The allocation and management of resources are crucial for wireless networks, in which the scarce wireless spectral resources are shared by multiple users. This requires efficient resource sharing mechanisms able to provide acceptable quality of service to each user, that depend strongly on both the amount of data to be transmitted as well as the capacity of the physical link. In particular, in wireless networks, where channel quality can vary dramatically in both time and frequency, a variety of link adaptation techniques such as adaptive modulation and coding or power allocation are employed that enable a user’s data rate to be adapted to the channel status. In addition channel-aware scheduling strategies are used to adaptively transmit data and dynamically assign wireless resources based on channel state information. The key idea is to choose one user with good channel conditions to transmit packets taking advantage of the multiuser diversity, whose gain increases with the number of users. Resource utilization can be efficiently improved also resorting to cognitive technology thus allowing concurrent systems to share the same resources without interference. Indeed, cognitive capability refers to the ability of sensing the radio environment in order to identify those portions of the spectrum that are unused at a specific time or location. Then the system reconfigures itself to enable the communication to be dynamically adapted to the radio environment.

Schedule

Abstract

"The lectures are devoted to fundamentals and applications of complex-valued neural networks with multi-valued neurons. Due to the computational and theoretical advantages that processing in the complex domain offers over the real-valued domain, the area of complex-valued neural networks is one of fastest growing research areas in the neural network community. In addition, recent progress in pattern recognition, robotics, mathematical biosciences, brain-computer interface design has brought to light problems where nonlinearity, multidimensional data natures, uncertainty, and complexity play major roles – complex-valued neural networks are a natural model to account for these applications. The most important notion underlying the theory of complex-valued neural networks is that of the phase information. The multi-valued neuron (MVN) is a complex-valued neuron with the inputs and output located on the unit circle. MVN has a circular activation function, which depends only on phase and projects a weighted sum onto the unit circle. These specific properties determine many unique advantages of MVN. The most important of them are the ability of MVN to learn non-linearly separable input/output mappings without any network and simplicity of derivative-free learning, which is based on the error-correction rule. MVN-based complex-valued neural networks also have a number of unique advantages. MVN-based Hopfield neural networks have shown unique capabilities as associative memories. The multilayer feedforward neural network with multi-valued neurons (MLMVN) significantly outperforms a classical multilayer perceptron (MLP) and many kernel-based techniques including SVM in terms of generalization capability and the number of parameters employed. The following topics will be covered: 1. Complex-valued neurons vs. real-valued neurons: higher functionality, possibility to learn non-linearly separable input/output mappings using a single neuron, proper treatment of phase, multi-valued neuron and its error-correction learning algorithm. 2. Multilayer neural network with multi-valued neurons (MLMVN): derivative-free backpropagation learning based on the generalized error-correction learning rule, batch learning. 3. Applications of MLMVN: solving various classification, pattern recognition, and time series prediction problems. 4. Intelligent image and signal filtering using MLMVN. 5. Image recognition using MLMVN-based analysis of frequency domain features."

Schedule

Abstract

'System performance is affected by a number of uncertainties: the input data, the state of computing devices (e.g. cache content), contention over the interconnection, etc. How should the uncertainties be accounted for? After establishing a connection with other approaches (benchmarking, stochastic analysis, etc.), the "real-time" perspective to the management of uncertainties will be illustrated. We will review the main techniques to analyze and design real-time systems. Finally, we will illustrate what the real-time research can offer to the recent challenges in cloud computing, IoT and cyber-physical systems.'

Schedule

Abstract

"As machine learning algorithms become more and more complex, selecting model's hyperparameters become more and more challenging. At least three different types of hyperparameters play a crucial role: optimization constants (such as learning rate, momentum, etc), loss/regularization constants (e.g. amount of L2 penalty or task similarity in multi-task learning), and hyperparameters that control the complexity of the hypothesis space (kernel constants, number/size of layers in deep learning, etc.). This course provides an introduction to Bayesian optimization approaches, sequential model-based techniques, and gradient-based methods. Hyperparameter optimization is also related to learning-to-learn and learning-to-optimize, two recent areas of research that have recently received significant attention. Some links between these areas will be explored in this course."

Schedule

Abstract

The course will present the methodologies, models and mathematical tools typically used to analyse large-scale social networks. It will then discuss the main results obtained in the research literature to analyse real large-scale social networks using the Social Network Analysis methodology. It will focus primarily on two abstractions commonly used to analyse Online Social Networks, i.e., the social and the interaction graph. The course will present and discuss the differences emerging in the OSN analysis when using these abstractions. Finally, it will discuss examples of micro-level analysis of the interaction graph, and how the highlighted properties at the micro scale impact on macro-level properties at the macro scale, such as for example, information diffusion patterns.

Schedule

Abstract

In the engineering of non-functional requirements, verification of quantitative properties of stochastic models enables early assessment of design choices and provides a model driven guidance for implementation and integration stages, with notable impact in the development of systems exposed to the intertwined effects of concurrency and stochastic durations. To this end, probabilistic model checking supports a systematic practice through which the same model can be verified against multiple probabilistic properties specified in some well defined language, able to analyze the impact on quality of different patterns of behavior, and open to automated regression verification when the model evolves. A number of techniques and tools have been proposed, based on statistical discrete event simulation or on numerical solution, the latter aiming at accuracy and confidence at the expense of restrictions on the class of models amenable to verification. In particular, if all model durations are exponentially distributed (EXP), the underlying stochastic process of the model falls in the class of Continuous Time Markov Chains (CTMC), the model always satisfies the Markov condition, and an efficient numerical solution can be attained by composition of behaviors [BHHK03,CHKM09,DHS09]. However, model validity often requires that some durations break the EXP memoryless property and be rather generally distributed (GEN). For instance, this occurs in aging processes accumulating memory over time, or in real-time systems or network protocols where correctness depends on firm time bounds. In these cases, the future evolution depends on the remaining time of GEN durations, and the underlying stochastic process may satisfy the Markov condition only at some regeneration points [BT95,CGL94,CKT94]. In this course we recall the basics of modeling systems using stochastic Petri Nets and illustrate their construction through the Oris tool [Oris], and we characterize different classes of stochastic processes that may underly a model depending on its structure, with emphasis on Continuous Time Markov Chains (CTMC), Semi Markov Processes (SMP), and Markov Regenerative Processes (MRP) [BT95,CGL94,CKT94]; we illustrate through examples the motivations for numerical solution of models with concurrent stochastic non-Markovian durations, focusing on cases with multiple concurrent GEN durations with an underlying stochastic Markov Regenerative Process (MRP); we describe the salient traits of numerical solution of reachability analysis for this class of models through the method of stochastic state classes [CGV09,VSC09,HPRV12]; we describe a novel technique that extends the method to solve a problem of probabilistic model checking emphasizing challenges, achieved results, and structural limits that arise when non-Markovian regenerative analysis is cast in the shape of probabilistic model checking [HPRV11,PHV15].

Schedule

Abstract

Schedule

Abstract

Schedule

Abstract

Schedule

Abstract

A fundamental limitation of standard machine learning methods is the cost incurred by the preparation of the large training samples required for good generalization. A potential remedy is offered by multi-task learning: in many cases, while individual sample sizes are rather small, there are samples to represent a large number of learning tasks, which share some constraining or generative property. If this property is sufficiently simple it should allow for better learning of the individual tasks despite their small individual sample sizes. The course review a wide class of multi-task learning methods, describe techniques to solve the underlying optimization problems and present an analysis of the generalization performance of these learning methods which provides a proof of the superiority of multi-task learning under specific conditions. We also highlight the link between MTL and learning-to-learn (or meta-learning) and reviews recent efforts in this direction.

Schedule

Abstract

The course deals with methods for the solution of smooth nonlinear optimization problems by assuming that the derivatives of the objective function can be neither calculated nor approximated explicitly. The interest in studying minimization algorithms for solving these optimization problems derives from the increasing demand from industrial and scientific applications for such tools. Derivative-free methods both for unconstrained and constrained optimization will be presented. As regards unconstrained problems, coordinate search methods and their variants will be analyzed. Concerning constrained optimitazion, the cases of box constraints, linear constraints and nonlinear constraints will be considered. Finally, derivative-free algorithms for the solution of nonlinear equations will be presented.

Schedule

Abstract

In the engineering of non-functional requirements of a large class of software, hardware, or physical systems, quantitative evaluation of stochastic models enables early assessment of design choices, and provides model driven guidance for implementation and integration stages. In particular, this becomes essential when facing the intertwined effects of concurrency and stochastic timing of systems that are supposed to meet real time requirements. Most relevant results in the literature of quantitative evaluation have addressed the verification of real-time properties over models with exponentially distributed (EXP) durations. In this case, the underlying stochastic process of the model is a continuous-time Markov chain (CTMC) enabling efficient and mature analysis algorithms. Yet, most systems subject to real time requirement natively include firm timing mechanisms (e.g. timeouts, watchdogs, periodic releases, bounded execution times) that break the assumptions of memoryless and unbounded behavior of EXP variables, and rather require representation of generally distributed (GEN) durations. In this case, the underlying stochastic process of the model falls in more complex classes of so-called non-Markovian processes, for which numerical solution is still to a large extent an open problem. In this talk, we address the problem of numerical solution of models with concurrent GEN durations, we outline the salient traits of the method of stochastic state classes which significantly enlarges the class of models amenable to numerical solution, and we report on examples of application based on the implementation provided by the Oris tool.

Schedule

Abstract

In recent years, network dynamics has emerged as a powerful paradigm in science and engineering. In fact, many natural and artificial systems can be modeled as networks of interacting multiple agents, in which the agents need to cooperate with each other in order to achieve global objectives. Examples of such multi-agent systems range from swarms and flocks in biology, to computer networks, social networks, multi-robot systems, sensor networks, vehicle platoons, power networks, etc. This course aims at providing the fundamental tools for modeling, analyzing, and designing the behavior of multiple agent systems, focusing on scalable and distributed (peer-to-peer) algorithms able to deal with large-scale systems and big data.

Schedule

Abstract

"The course aims at introducing some instruments for performance modelling by means of queueing systems and queueing networks. Particular attention will be devoted to the analysis of case studies taken from the literature of software performance engineering. Specific topics addressed by the course: Basic queueing systems - Structure of a queueing system and Kendall's notation - Fundamental results on queueing systems (M/M/1, M/M/k, M/M/Infinity, M/G/1) - The processor sharing discipline and the M/G/1/PS: the insensitivity property - Optimality of the Shortest Remaining Processing Time discipline Queueing networks - Basic definitions - Operational Analysis - Burke's theorem and feed-forward networks - Jackson's and Gordon Newell's queueing networks - PASTA and Arrival Theorem - Algorithms for closed queueing networks - Case studies and new research challenges"

Schedule

Abstract

"Based on the book Machine Learning: A Constrained-Based Approach"

Schedule

Abstract

In this course we see how to build systems for recognizing the semantic content of images and videos. We start with the historical context given by first attempts by Marr and colleagues to formalize recognition of visual phenomena, then proceed to develop the Bag-of-Features models of visual recognition that led to the first significant breakthroughs in object recognition, and finally to the Convolutional Neural Network (CNN) models that are the current state-or-the-art. We will finish with an overview of recent work following the trend toward unsupervised, generative models like the GAN and the Variational Autoencoder. The course is organized around assigned readings of seminal and recent papers in the field.

Schedule

Abstract

The course will introduce concepts related to Quality of Service in packet networks, with particular reference to scheduling algorithms. We will show how to derive worst-case delay guarantees for a generic packet scheduler, and then we will show how Network Calculus allows one to compute bounds on the delay, the backlog, and the output for a flow traversing a multi-node path.

Schedule

Abstract

Optimization methodology is the engine of prescriptive analytics. This short course gives a gentle, rigorous introduction to modern convex optimization models and algorithms. Duality provides optimality conditions and serve as the platform of algorithm design. First we focus on duality in linear optimization (LO) and convex conic linear optimization (CLO) problems. CLO includes the LO, second order conic and semidefinite optimization problems, which are solvable by Interior Point Methods (IPMs) in polynomial time, and also the NP-hard classes of copositive and completely positive CLO problems. Robust LO models will motivate the introduction of second order conic optimization problems. Then algorithmic concepts, such as pivot algorithms and interior point methods (IPMs) are discussed. Initialization of the algorithms, the computational cost and efficient computation of an iterative step, and characteristics of the produced optimal solutions are discussed. As time allows we shortly available software packages, sensitivity analysis, and some applications will be discussed. Note: Preliminary knowledge of optimization, operations research models and methods is a plus, but everyone with good linear algebra and multi-dimensional calculus skills should be able to follow the course.

Schedule

Abstract

The problem of coordinating multiple UAVs for distributed targets detection and tracking is modelled as the adaptation problem of a team of UAVs to mission-specific environmental layouts. Biological mechanisms of flocking and stigmergy are the basis of the formation of a swarm, while the adaptation is based on an evolutionary algorithm with the objective of maximizing the number of tracked targets during a mission or minimizing the time for target discovery.

Schedule

Abstract

"The course presents the basic cryptographic primitives: ciphers, hash functions, and digital signatures. Particular attention will be placed on their security and performance. Real applications will be discussed."

Schedule

Abstract

TBA

Schedule

Abstract

None

Schedule

Abstract

"This course will introduce theory of artificial neural networks. The theoretical properties of networks will be discussed focusing on how those properties affect practical applications. The purpose is that of providing the students with the basic background to better comprehend the motivations that determinate the common practice in neural network applications. The considered architectures will include the common architectures, multilayer neural networks, recurrent neural networks, deep learning. A particular attention will be also dedicated to graph neural networks."

Schedule

Abstract

"The course aims ​to ​provid​e​ an introduction ​to ​graph-based clustering principles and methods, and to discuss some applications​ to selected computer vision problems.​ The student is expected to be familiar with the basic concepts ​of calculus and linear algebra. Course contents 1. Introduction (1 h) 2. Cut-based methods. Normalized-cut and spectral clustering (4 h) 3. Clique-based methods. Dominant-set clustering and its variants (4 h) 4. Selected computer vision applications/ image segmentation, group behavior analysis, geo-localization, multi-target tracking, person re-identification (3 h)"

Schedule

Abstract

"- This course is about using stochastic Petri Nets to specify the probability space of some system with concurrent stochastic durations; - ... and then solving stochastic processes identified in that probability space to evaluate measures of interest about qualities of the system, e.g. about performance or dependability. - The approach is first developed in a Markovian setting with exponentially distributed durations, using the formalism of Generalized Stochastic Petri Nets (GSPN), whose underlying stochastic process falls in the class of Continuous Time Markov Chains (CTMC) and Discrete Time Markov Chains (DTMC); - ... and it is then extended to a non-Markovian setting, where models also include generally distributed durations and thus identify an underlying Markov Regenerative Process (MRP); - ... with a balance between ground and advanced concepts that will be adapted to fit the composition of the class. - The Oris tool and its Sirio API (https://www.oris-tool.org/) will be presented, reporting on cases of application and providing means for hands-on experience in modeling and evaluation."

Schedule

Abstract

"This doctoral course will present advanced system modelling techniques in use to evaluate performance and scalability in software systems and services. The module will alternate theoretical foundations, case studies, and hands-on experience. The syllabus touches upon three main topics: (1) stochastic modelling of software systems and service workflows; (2) model evaluation techniques; (3) learning model parameters based on production data. The module will include demos and hands-on exercises with Java Modelling Tools (http://jmt.sf.net/), a family of graphical tools for simulation and analytical evaluation of complex system models. The module does not assume prior knowledge of modelling and is open to PhD students from all technical backgrounds."

Schedule

Abstract

"Specialized accelerators are the key for energy efficiency for all future systems. While fixed-function accelerators are the most efficient, the use of reconfigurable accelerators allows the in-field and runtime specialization of functions, combining hardware efficiency with broad flexibility. This course will cover the basics of using reconfigurable technology (FPGAs), and then proceed to describe how it can be used for function accelerator in heterogeneous systems. We will cover full and partial reconfiguration, the necessary support they require, and their limitations. We will also discuss the pros and cons of a spectrum of commercial and research choices for implementing accelerators focusing on the interconnection/communication with the host processor and memory. A hands-on lab will focus on the creation of simple accelerators and linking them to simple applications."

Schedule

Abstract

"This course gives an introduction to probabilistic graphical models. Various important types of graphical models are introduced, and their use in typical application domains illustrated (e.g., Hidden Markov Models for biological sequence analysis, Markov Random Fields for Image Processing). The course emphasizes both modeling and learning aspects of graphical models. With regard to modeling, we discuss underlying probabilistic (independence) assumptions that need to be made when choosing a particular type of model for a given application domain. Also limitations imposed by the computational complexity of inference tasks for a given model are considered. With regard to learning, the most important paradigms for learning graphical models from data are explained, and exemplified by typical supervised and unsupervised learning tasks. Outline of contents: Basics of probability: - joint probability distributions - conditional probability and independence Directed and undirected graphical models - Bayesian and Markov networks, factor graphs - Graph structure and independence relations: d-separation and the Hammersley-Clifford theorem Learning - Structure learning - Parameter estimation - Supervised learning Latent variable models - Hidden Markov models - Unsupervised learning Bayesian approaches - Bayesian learning - Plate representation - Nonparametric Bayesian models Beyond graph-based representations - template models - logic-based representations - statistical relational learning"

Schedule

Abstract

"The course will review some deep architectures for sequence learning and graph learning, including recurrent, convolutional and attention-based approaches"

Schedule

Abstract

"Today's cryptography is standing on a brink of a revolution, which somehow resembles the one happened in the '70s that introduced public-key cryptography with the seminal Diffie and Hellman paper "New Directions in Cryptography." This course will focus on some new crypto technologies that promises to revolutionize our world: from blockchain to attribute-based encryption, from post-quantum cryptography to LWE cryptosystems. TOPICS: - Zero-Knowledge Proofs, Non-interactive Zero-Knowledge Proofs (NIZK). - Blockchains, proof-of-work, proof-of-stake. - Cryptography proofs. - Introduction to elliptic-curve cryptography. - Pairing-based cryptography. - Identity-based encryption, attribute-based encryption. - Puncturable encryption. - Introduction to post-quantum cryptography. - Lattice-based Cryptography, lattice problems, q-ary lattices. - NTRU cryptosystem. - Learning-with-errors (LWE) cryptosystems."

Schedule

Abstract

"The course we will review some approaches to interpretability in machine learning."

Schedule

Abstract

Software Product Line Engineering (SPLE) is a software engineering methodology aimed at developing, in a cost-effective and time-efficient manner, a family of highly configurable (software-intensive) systems through systematic reuse. Individual variants (called products) share an overall family model, but they differ with respect to specific features, which are (user-visible) increments in functionality. The explicit introduction and management of feature-based variability in the software development cycle causes complexity in the modelling and analysis of such highly configurable systems, since the number of possible variants may be exponential in the number of features. Recent research has focussed on lifting formal specification languages and formal verification techniques known to be successful for single (software) systems, such as process calculi and model checking, to (software) product lines or highly configurable systems in general. In this course, after a general introduction to SPLE and formal modelling and analysis techniques, in particular model checking, we present a number of modelling and analysis frameworks that have been introduced for the specification and verification of variability in behavioural models of SPLs. These range from variants of Modal Transition Systems (MTSs) and Featured Transitions Systems (FTSs) with (action-based, feature-based) CTL logics and their dedicated family-based model checkers VMC and SNIP to a feature mu-calculus over FTSs that allows family-based model checking with off-the-shelf model checkers like mCRL2. We will also consider rich domain-specific languages for SPLE like QFLan, with applications to the security domain thanks to RisQFLan, QFLan's instantiation to such domain. These languages are equipped with simulation-based analysis, and in particular statistical model checking through the analyzer MultiVeStA, which we introduce.

Schedule

Abstract

The development of software in safety-critical domains is subject to specific guidelines that aim at reducing the possibility that bugs in the software threat safety of the users. The course will introduce the software safety guidelines adopted in three transportation domains, namely railway signailling, avionics, automotive, and will discuss the differences among them and their actual implications on the software development process.

Schedule

Abstract

"Abstract: Research on Deep Learning techniques has mainly focused so far on data defined on Euclidean domains (i.e., grids). However, in a multitude of different fields, such as Biology, Physics, Network Science, Recommender Systems and Computer Graphics, one may have to deal with data defined on non-Euclidean domains (i.e., graphs and manifolds). The adoption of Deep Learning in these fields has been lagging until very recently, primarily since the non-Euclidean nature of data makes the definition of basic operations (such as convolution) rather elusive. Geometric Deep Learning deals with the extension of DL techniques to graph/manifold structured data. In this course, we will introduce this area of research by presenting recent research advancements in DL applied to point clouds, graphs, meshes and manifolds. Example applications will be illustrated in laboratory sessions."

Schedule

Abstract

Schedule

Abstract

Schedule

Abstract

"The course introduces the use of formal methods for the specification and verification of quantitative properties of concurrent distributed systems. In particular we look at the foundations of such techniques among which various kinds of Markov chains and related probabilistic and stochastic process algebras for the modelling of systems, and to probabilistic and stochastic temporal logics for the specification of properties. Furthermore, the basic probabilistic and stochastic model-checking algorithms will be illustrated. Hands-on experience will be provided by performing smaller case studies in modelling and verification using the state-of-the-art PRISM model checker. Some lectures will be dedicated to illustrate more advanced recent developments and perspectives for future research in this field in particular in relation to the modelling and verification of smart and collective systems."

Schedule

Abstract

Empirical studies in software engineering provide a systematic way of evaluating theories, languages, concepts, tools or methodologies, considering the industrial context in which they are applied [1]. The course will prepare students by examining how to plan, conduct and report on empirical studies in software engineering. The course will cover all of the principal methods applicable to software engineering (controlled experiment, case studies, surveys, systematic literature reviews, and ethnography) and will describe quantitative and qualitative methods of analysis, including hypothesis testing and grounded theory. To showcase the different methods, the course will critically review representative examples of published work. At the end of the course, the students will be able to approach real-world research problems in a scientifically sound way, and contribute to theory building in software engineering research. A previous version of this course was delivered for the MSc and Ph.D students of the University of Florence, School of Mathematical, Physical and Natural Sciences. The web version of the course has been made available on YouTube [2]. The reference book for the course is the handbook from Wohlin et al. [1]. Course Contents in brief: • Overview of Empirical Methods in Software Engineering Research • Formulating Research Questions • Data Types, Measurements, Scale • Data Collection Techniques • Building Theories in Software Engineering • Research Strategies: the ABC Framework [3] • Controlled Experiments • Hypothesis Testing and Statistical Tests • Qualitative Research Methods: Ethnography, Interviews, Grounded Theory • Survey Research in Software Engineering • Case Studies • Systematic Literature Reviews References: [1] Wohlin, C., Runeson, P., Höst, M., Ohlsson, M. C., Regnell, B., & Wesslén, A. (2012). Experimentation in software engineering. Springer Science & Business Media. [2] A. Ferrari, Empirical Software Engineering Lectures (2020). https://www.youtube.com/playlist?list=PLSKM4VZcJjV-P3fFJYMu2OhlTjEr9Bjl0 [3] Stol, K. J., & Fitzgerald, B. (2018). The ABC of software engineering research. ACM Transactions on Software Engineering and Methodology (TOSEM), 27(3), 1-51.

Schedule

Abstract

"This course gives a background on multi-objective optimization (MOO) along with the knowledge about developing and tuning efficient multi-objective genetic algorithms (MOGAs) for optimization problems of science and engineering. The course starts with a mathematical background on MOO. The main concepts of evolutionary computation and genetic algorithms are then introduced for real, integer, mixed-integer, and combinatorial optimization problems. Some techniques to handle constraints in MOGAs are also presented. The focus then moves on some approaches to find the best parameter values, and to carry out performance evaluation. Some multi-criteria decision making and data visualization techniques complete the set of topics covered by the course, with particular reference to choosing the final Pareto-optimal solution to implement."