Past seminars:

• 3 November 2015
  Speaker: Sara Ugolini.
  Title: A tour through Mathematical Logic towards Fuzzy Logic and the probability of vague events
  Abstract: Mathematical logic studies correct reasoning, and the idea around which every logic evolves is the concept of truth, and the way it spreads through the propositions. We will see the different concepts of truth from which respectively classical logic, intuitionistic logic and fuzzy logic were born, exploring foundational issues. Then we will see how universal algebra becomes a perfect setting to study all of these logics. In the end, we will focus on how it is possible to relate fuzzy logic to probability theory, always with an algebraic flavour, dealing in particular with the subjective foundation and with the concept of many-valued events.
• 17 November 2015
  Speaker: Alessio Conte.
  Title: Enumerating patterns in real-world graphs
  Abstract: Graphs are ubiquitous: the world wide web, social networks, traffic networks, computer networks biological networks... Great effort has been spent trying to mine information from real world graphs, but the task is challenging due to the linked nature of graphs, and the size that real world data can reach. Enumerating patterns in graphs is a way to discover high quality information, as the presence of different patterns can reflect the different dynamics involved in the structure of a network. This allows deeper understanding of the data and is a starting point for operations such as clustering and classification. Making enumeration feasible on real world networks requires a careful analysis of the problem and non-trivial novel ideas.
• 24 November 2015
  Speaker: Fabrizio Luccio.
  Title: Arithmetic for Rooted Trees
  Abstract: We propose a new arithmetic for rooted unordered trees of n vertices and a method for their enumeration. We define the operations of addition, addition-plus, and multiplication, present their properties of associativity and commutativity, and show that all trees can be generated by addition and addition-plus from a starting empty tree. We also show that some trees cannot be obtained as the sum, sum-plus, or product of two trees, thus defining prime trees with respect to the three operations, and prove that primality can be decided in time polynomial in n.
• 15 December 2015
  Speaker: Giuseppe Attardi.
  Title: The tsunami of Deep Learning over NLP
  Abstract: Deep Learning (DL) has emerged as a major breakthrough for machine learning (ML), combining unsupervised learning from large amounts of data with supervised learning for specializing learning to a given task. Unsupervised learning of distributional semantic representations frees ML system development from the tyranny of feature engineering. Building a DL system encompasses designing a model architecture and a top-level loss function; then both the parameters and the representation of the model self-organize as to minimize this loss, in an end-to-end learning framework. I will survey a number of natural language (NL) applications where DL has been successful: these include traditional NLP tasks such as POS tagging, NER tagging, parsing, translation, sentiment analysis as well as new applications such as Image Captioning, AdWords, Named Entity linking, sentence compression. Advanced DL libraries provide means to exploit parallelism both at the processor level (through GPUs) and at the cluster level. These include Theano, Caffe, Keras, TensorFlow. DeepNL is a DL library specific for NLP. Most of these exploit code generation techniques to produce efficient parallelized code. DL has also the potential to eliminate the dichotomy in question answering systems between the acquisition phase (ML based) and the deductive phase (based on logic inference). A recursive neural network (RNN) exploiting Long Short-Term Memories (LSTM) can be used to do both.
• 1 March 2016
  Speaker: Marco Grandi.
  Title: Covariance and Contravariance in Programming Languages
  Abstract: The concepts of covariance and contravariance in programming languages are related to the notion of subtyping. In a nutshell, these terms refer to the fact that a type constructor preserves or reverses the subtype relation, which is a preorder on types. Firstly, I present these concepts from a general point of view and focus on a delicate type constructor, namely the functional constructor. Then, I speak of these terms in the setting of static typed object-oriented languages, where subtyping is commonly associated with inheritance. I also briefly remark the two different approaches for method specialization: contravariant specialization and covariant specialization. The latter is adopted in the Eiffel programming language. Finally, I explain how variance can be combined with generics and show as example a popular programming language, i.e. the Scala programming language.
• 26 April 2016
  Speaker: Massimo Torquati
  Title: FastFlow, a programming model for fast Data Streaming on heterogeneous multi-cores
  Abstract: In recent years, we have seen an explosion of data streaming coming from the huge number of ubiquitous devices. Traditional big data management systems are designed for high throughput of batch jobs; they can hardly deal with urgency and latency constraints. Moreover, modern applications ask for elasticity and self-adaptivity features. They must be gracefully and dynamically adaptable to match real context needs. In the setting of the Internet of (Every)Thing and Fog computing, applications cannot be statically configured to sustain peak loads, this is simply too costly both in terms of resources needed and power consumed. To tackle all these problems, we need suitable abstractions for application developers. Domain-specific experts need to concentrate on developing newer and better applications and do not have to spend their time in writing highly-tuned low-level code. In this context, the FastFlow parallel programming framework may provide not only some powerful mechanisms, but also a methodological approach. FastFlow has been originally designed having in mind three keywords: streaming, performance and flexibility. During the talk, we present the main features of FastFlow showing also how it can be used to parallelize simple programs.
• 3 May 2016
  Speaker: Elena Salvatori
  Title: Data analytics for educational processes: a case study for assessing foundation skills of adult students
  Abstract: As a result of the massive introduction of ICT, education will face new challenges in the next future such as rethinking the role of teachers, the growing ubiquity of social media and the integration of hybrid learning. Open Educational Resources, hybrid learning design, data-driven learning assessment, the evolution of online learning and the acceleration of intuitive technologies will also play an important role. The speed of accessing ICT in today's world as well as the changes occurred in the social, economic and cultural landscapes require citizens to master an expanded set of skills to cope with new media. The OECD, the EU and the Italian government, recognizing the strategic importance of human capital for economic growth, have already launched large programs to tackle the ICT's challenges faced by educational institutions. But how can we take advantage from the huge amount of data collected so far? How can we measure the "economic" return of foundation skills? And also, what do we know about the return of the massive investments in ICT in terms ICT skills? The talk will start introducing the research field of data analytics applied to educational data and a selection of case studies published in this research field. This will be followed by a section of the 21st skills' strategies put in place by EU, OECD and Italian institutions. Finally, a case study on assessing foundation skills of adult students in the province of Grosseto will be shown.
• 10 May 2016
  Speaker: Manuele Sabbadin
  Title: Global illumination of raw scanning data
  Abstract: Ambient Occlusion (AO) is a well known technique that is used to generate effective approximation of global illumination effects in rendering. Its impact in improving the readability and understandability of three-dimensional shapes is well known. However the use of this visual enhancing technique in the management of raw scanning and acquisition data remained very limited. One of the main issues that prevents the use of AO in these context is the efficient computation of the global illumination effects of the scene. Since this computation is hard and complex, several real-time methods have been proposed to approximate these global effects by using simplifying assumptions. One of these methods, named Screen Space Ambient Occlusion, is well known in the field of real time renderings. Since the nature of the scanned data makes hard to apply this algorithm to them, a new algorithm has been made, called MultiViewAO. The lecture will cover an introduction on lighting and 3D scanning, then two algorithms will be presented: MultiViewAO and a real-time extension used to improve data captured by a Kinect.
• 12 May 2016
  Speaker: Giulio Masetti
  Title: Impact of Equations Ordering on Inexact and Quasi-Inverse Newton Methods
  Abstract: The Newton-Raphson method has a prominent role in applied science and engineering. Our research group at SEDC lab., ISTI/CNR, is involved in dependability-related assessment of Smart Electrical Grids and the Newton-Raphson method is used to solve numerically the system of power-flow equations that characterize the electrical network. In general, given a system of non-linear equations written in the form F(u)=0, where F is a continuously differentiable function that maps a n-dimensional vector u to a n-dimensional vector F(u), one of the solutions can be found via the Newton-Raphson method. At each iteration of this method, a linear system that involves the Jacobian matrix J_{F}(u) of F has to be solved and the equation ordering does not affect it. At increasing the size of the problem, solving with a direct method the linear system is often impractical (w.r.t. time and space complexity, and accuracy), even if the Jacobian is a sparse matrix. Therefore, in these cases iterative methods have been studied, where, the equation ordering becomes a critical aspect to analyse carefully. Studies related to equations ordering are available on techniques employed in medium-size linear systems, such as the Gauss-Seidel method. In this method, at each iteration, permutations of Jacobian rows and columns, that do not depend on the initial equation ordering, are exploited to gain in CPU time. Jacobian-free techniques, developed to solve very large linear problems, offer interesting opportunities for investigations on the impact of initial equations ordering over the number of iterations. One of the most effective algorithms in this category is the Inexact-Newton-Krylov: it gets rid of the Jacobian matrix and approximates the matrix-vector product J_{F}(u)v with the finite difference [F(u+sv)-F(u)]/s. Another variant of Newton- Raphson is the Quasi-Inverse-Newton: it approximates the inverse of the Jacobian J_{F}^{-1}(u) with a diagonal matrix and solves directly the linear system. A seminal article, written by L. Dutto and published on the International Journal for Numerical Methods in Engineering in 1993, numerically presents the influence of initial ordering over the preconditioner of Inexact- Newton-Krylov GMRES, and then on the number of iterations. However, it does not address the impact of ordering on the core of the method, that is the adopted Jacobian-free techniques. To go further such existing results, our study focuses on Jacobian- free methods and explores the influence of the initial ordering over them. In this talk, we discuss our approach to prove that the initial equations ordering affects Inexact-Newton-Krylov and Quasi- Inverse-Newton, also presenting a preliminary analysis to quantify the resulting impact. A trivial bivariate function will be used as case study and numerical experiments will be presented. Interestingly, the equations ordering does not only affect the efficiency in reaching a solution, but also the solution itself. As next step, we are working on exploring more realistic scenarios in the power grid domain. Traditionally, the characteristics of the domain at hand has influenced the application of mathematical structures. For example, when using Gauss-Seidel in the power-flow context, there exists a particular initial ordering such that the Jacobian is a four-block matrix and each block corresponds to well defined electrical quantities. Using Jacobian-free methods the application-oriented initial equations ordering could not be the best choice with respect to iterations number and reachable solutions. We believe this is an important feature, that the user of these methods need to be aware of.
• 17 May 2016
  Speaker: Massimiliano Bertolucci
  Title: Static and Dynamic Big Data Partitioning on Apache Spark
  Abstract: Many of today's large datasets are organized as a graph. Due to their size it is often infeasible to process these graphs using a single machine. Therefore, many software frameworks and tools have been proposed to process graph on top of distributed infrastructures. This software is often bundled with generic data decomposition strategies that are not optimised for specific algorithms. In this paper we study how a specific data partitioning strategy affects the performances of graph algorithms executing on Apache Spark. To this end, we implemented different graph algorithms and we compared their performances using a naive partitioning solution against more elaborate strategies, both static and dynamic.
• 24 May 2016
  Speaker: Laura Pollacci
  Title: Rap: more than words
  Abstract: Musical genres can provide a representative overview of many social and linguistic phenomena. This work aims to analyze the Italian rap music, since this genre provides the most interesting cues for a textual analysis. The main focus of the work is the development of a tool able to map Italian rap language and to offer an overview of the rap universe. The analysis is performed using Natural Language Processing techniques and data visualization tools, in order to highlight latent relationships, categories and relevant features of the rap network.
• 31 May 2016
  Speaker: Luca Pedrelli
  Title: Dynamic Neural Networks for Balance Assessment Estimation
  Abstract: The evaluation of balance assessment in elderly people is a relevant aspect in the field of health monitoring. The Berg Balance Scale (BBS) represents a golden standard test for clinical assessment of balance stability. Recently the use of Wii Balance Board (WBB) is proposed from literature as valid alternative to clinical test of balance assessment. In our work we process the whole multidimensional temporal signal generated from the 4 WBB sensors by means of dynamical neural networks. In particular, in the context of DOREMI European project (GA611650), we deploy an automatic test, based on a Recurrent Neural Networks (RNN) system, to estimate BBS score of the patients.
• 9 June 2016
  Speaker: Farzad Vaziri
  Title: Mobility analysis on real and simulated data
  Abstract: All electronic devices are making data and in first glance the data are made by them probably does not look very interesting. But if we look a bit forward and change our perspective we can understand that all theseuseless data can be converted to useful informations. Electronic devices are getting more and more popular among the people and most interesting ones are portable devices which actually need to work wireless and it makes very potential analyzing the data made by them, because we have chance to use trajectories and trace of users behavior. One of the most popular works which a lot of studies have done about it is traffic control managements, either for cars or pedestrians.
• 21 June 2016
  Speaker: Giulio Ermanno Pibiri
  Title: Elias-Fano Encoding
  Abstract: The so-called succinct data structures have acquainted a lot of attention in recent years for their double promising goal: compress data with performance close to the information theoretic lower bound while supporting exceptionally fast access to data paying a negligible, lower-order term, space factor. This talk presents an overview of the Elias-Fano encoding of monotone integer sequences, with particular attention to the killer applications where the encoding has proven to shine.
• 28 June 2016
  Speaker: Giovanna Broccia
  Title: Model Checking, an overview of new application areas
  Abstract: Model checking is a formal verification technique which allows for desired behavioral properties of a given system to be verified on the basis of a suitable model of the system through systematic inspection of all states of the model. In the last two decades these techniques have considerably developed, especially in the field of the Computer Science as a method to verify the absence of error and for debugging. However it is not hard to understand how these techniques can help other field of study and how the construction of model ​ad hoc and the formal verification of desired properties is a useful technique in different field of knowledge. This talk presents an overview of some new application areas.
• 5 July 2016 2016
  Speaker: Francesca Pratesi
  Title: Mauriana Pesaresi Seminars: a meta seminar
  Abstract: This seminar will be a sort of summary of the last three cycles of Mauriana Pesaresi Seminars. I will show some simple statistics, and I will describe you lights and shadows of being the organizer of a PhD Seminar. Anyway, don't worry: it will be a very short seminar!