Data e Ora : Thursday, December 10, 2015 - 14:30
Relatore : Francesca A. Lisi
Affiliazione : Dipartimento di Informatica dell’Università degli Studi di Bari "Aldo Moro"
Luogo : Aula Magna A. Lepschy
Descrizione :
Abstract: I progressi recenti nell'area di ricerca nota come "Computational Creativity" (CC) testimoniano un interesse crescente nell'uso di calcolatori per generare risultati che sarebbero considerati creativi se ottenuti da esseri umani da soli. D'altro canto, costruire macchine creative Ë anche considerato un traguardo per la ricerca in Intelligenza Artificiale (IA). Tuttavia, ricerca fondamentale in IA dovrebbe essere svolta per renderla capace di affrontare le sfide della CC. In particolare, il test di Turing si rivela inadeguato nel caso di macchine creative. In questo seminario, illustrerò brevemente il Lovelace Test, intitolato alla memoria di Ada Byron Lovelace e proposto come strumento di valutazione della creatività nelle macchine. Nel bicentenario della nascita di ADA BYRON LOVELACE
Data e Ora : Wednesday, December 2, 2015 - 16:30
Relatore : Prof. Paolo Zanotti Fregonara
Affiliazione : Università di Bordeaux, Francia
Luogo : Sala Riunioni DEI/G N. 318 III piano
Descrizione :
Abstract: Per studi PET (Positron Emission Tomomgraphy) dinamici avanzati quali lo studio dell’infiammazione con il tracciante radiomarcato [11C]PBR28 è fondamentale l’acquisizione di campioni arteriali per descrivere la funzione di ingresso da usare nel processo di quantificazione.La necessità di prelevare campioni arteriali rende, però, l’esperimento PET difficilmente applicabile in ambito clinico. Per superare questa limitazione, in letteratura sono presenti diversi metodi alternativi quali Image DerivedInput Function o Population Based Input Function che, sotto determinati vincoli, permettono di evitare il campionamento arteriale. Questa presentazione coprirà i principali metodi proposti, descrivendo i loro vantaggi e le difficoltà metodologiche.
Data e Ora : Friday, November 6, 2015 - 14:30
Relatore : Prof. Fabrizio Luccio
Affiliazione : Università di Pisa
Luogo : Sala Riunioni 318 DEI/G
Descrizione :
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 timepolynomial in n. We suggest how these concepts can be useful and discuss related studies appeared in the literature.Arithmetic for rooted unordered trees is completely new. Some open problems arise, other tree operations could be proposed, and our results could be improved. Suggestions and comments will be warmly welcome. Reference:arXiv:1510.05512v1
Data e Ora : Tuesday, November 3, 2015 - 12:00
Relatore : Dr. Marco Bressan
Affiliazione :
Luogo : Sala L. Merigliano DII-Sede G
Descrizione :
How large a fraction of a graph must one explore to approximate a quantity that depends on its global structure? This question underpins many algorithmic problems arising in modern large-scale networks, where several factors limit the access to the underlying graph topology. In the case of graph centrality measures, considerable effort has been devoted to approximating the score of a node with limited information about the rest of the graph; but for PageRank, a classic random-walk based centrality measure, no general solution was known until now. In this talk we present the first algorithm that approximates the PageRank score of any node in any graph by exploring only a sublinear number of nodes. We show that our result is essentially optimal, and that any combination of exploration primitives and error guarantees other than the one we use makes it impossible to achieve sub-linearity. This result proves that, for a specific but nontrivial class of Markov chains, the stationary probability of any given state can be approximated exploring only a vanishing portion of the transition matrix. This is a joint work with Enoch Peserico and Luca Pretto.
Data e Ora : Friday, October 30, 2015 - 14:15
Relatore : Dott. Giulio Caravagna
Affiliazione : University of Edinburgh
Luogo : sala riunioni 318 DEI/G
Descrizione :
The evolutionary nature of cancer relates directly to a renewed focus on the voluminous NGS (next generation sequencing) data, aiming at the identification of explanatory models of how the (epi)genomic events are choreographed in cancer initiation and development. However, despite the increasing availability of multiple additional -omics data, this quest has been frustrated by various theoretical and technical hurdles, mostly related to the dramatic heterogeneity and temporality of the disease. In this paper, we build on our recent works on selectivity relation among driver mutations in cancer progression and investigate their applicability to the modeling problem - both at the population and individual levels. On one hand, we devise an optimal, versatile and modular pipeline to extract ensemble-level progression models from cross sectional sequenced cancer genomes. The pipeline combines state-of-the-art techniques for sample stratification, driver selection, identification of fitnes s-equivalent exclusive alterations and progression model inference. We demonstrate this pipeline's ability to reproduce much of the current knowledge on colorectal cancer progression, as well as to suggest novel experimentally verifiable hypotheses. On the other hand, we prove that our framework can be applied, mutatis mutandis, in reconstructing the evolutionary history of cancer clones in single patients, as illustrated by an example with multiple biopsy data from clear cell renal carcinomas.
Data e Ora : Wednesday, October 28, 2015 - 14:30
Relatore : Dr. Jari Kleimola
Affiliazione : Aalto University
Luogo : Aula Magna DEI
Descrizione :
Abstract:The advancements in web technologies have extended the sandbox of the web browser, and increased the usability of the web as an application platform. The browser multimedia stack was recently complemented with W3C Web Audio API standard, which enables audio synthesis and processing in real-time using JavaScript or cross-compiled C/C++ code. The code loads directly from the open web without manual plugin installations, and integrates seamlessly with the rest of the page content and other client-side APIs.This talk first describes the concepts and current state of the Web Audio API, points out its limitations, and discusses the on-going standardization efforts for their mitigation. The talk then explains how existing C/C++ codebase can be reused in browser environments using Emscripten and PNaCl technologies. Finally, I will demonstrate three emerging use cases for web audio: 1) Web Audio Modules, which are the equivalent of VST plugins in web browsers, 2) porting Pd vanilla to the web, and 3) the topic of my research visit at the SMC group related to individualized HRTFs in web browsers.
Data e Ora : Monday, October 5, 2015 - 10:30
Relatore : Prof. Jun Miura
Affiliazione : Department of Computer Science and Engineering, Toyohashi University of Technology, Japan
Luogo : Aula Magna Lepschy
Descrizione :
Abstract:This talk presents our recent projects on autonomous mobilerobots. The projects aim at developing personal service robots thatcan assist people in their daily life. The ability of movingautonomously in various environments, both indoor and outdoor, is oneof the fundamental functions for such robots, and is realized bycombining several technologies such as localization and mapping,motion planning, and human detection and tracking. The talk will coverthe following topics: development of person following robots, a newviewpoint planning strategy for an attendant robot (or guard robot),and multisensory fusion for reliable road detection.
Data e Ora : Wednesday, September 30, 2015 - 12:00
Relatore : prof. Pierluigi Crescenzi
Affiliazione : Universita' di Firenze
Luogo : Sala Riunioni 318 DEI/G
Descrizione :
Abstract: Closeness is an important centrality measure widely used in the analysis of real-world complex networks. In particular, the problem of selecting the k most central nodes with respect to this measure has been deeply analyzed in the last decade. However, even for not very large networks, this problem is computationally intractable in practice: indeed, Abboud et al have recently shown that its complexity is strictly related to the complexity of the All-Pairs Shortest Path (in short, APSP) problem, for which no subcubic ``combinatorial'' algorithm is known. In this paper, we propose a new algorithm for selecting the k most closeness central nodes in a graph. In practice, this algorithm significantly improves over the APSP approach, even though its worst-case time complexity is the same. For example, the algorithm is able to compute the top $k$ nodes in few dozens of seconds even when applied to real-world networks with millions of nodes and edges. We will also experimentally prove that our algorithm drastically outperforms the most recently designed algorithm, proposed by Olsen et al. Finally, we apply the new algorithm to the computation of the most central actors in the IMDB collaboration network, where two actors are linked if they played together in a movie. Joint work with Michele Borassi and Andrea Marino.
Data e Ora : Thursday, September 24, 2015 - 10:30
Relatore : Kahori Kita
Affiliazione : Assistant Professor at Center for Frontier Medical Engineering, Chiba University (Chiba, Japan).
Luogo : Aula Magna "A. Lepschy"
Descrizione :
Abstract: My research interests focus on rehabilitation engineering and include motor control/learning of healthy subjects as well as patients. During my talk I will introduce some key aspects of my research topics, especially concerning rehabilitation devices for stroke patients and the neural basis of musician’s dystonia. Sensory disturbance is common following stroke, and sensory loss may inhibit the ability of patients to use their hands in their daily activities. In other words, the “manipulation capability”, even though they have a good motor function, is compromised. Particularly, "motor function" is assessed by passive and active range of motion (spasticity problems included), while "manipulation capability" is the ability of grasping or pinching objects in daily activities. We hypothesized that sensory feedback training might enhance task-oriented training and improve manipulation capability. Therefore, we developed a novel sensory feedback system by electrical stimulation in which patients receive a sensory feedback during pinching or grasping objects. I will show the proposed system and results of a first feasibility test with stroke patients. On the other side, through long-term and excessive motor training, musicians may develop musician’s dystonia (MD). Typically in MD at the hand, the affected fingers become uncontrollable mostly during playing a specific musical instrument, but respond normally to other motor activities. Some studies recently found relations between cerebellum functioning and dystonia disease. For instance, patients with MD show less cerebellar activity during a bilateral finger-tapping task, and greater cerebellar activation is observed during writing in writer’s cramp (another form of task-specific focal dystonia). However, it remains unclear how the cerebellar abnormality results in abnormal muscle contraction in MD. We hypothesized that the cerebellar activity might functionally modulate activity of the motor-related areas, yielding abnormal muscle contraction during movements in MD. We have conducted MRI experiments to validate this hypothesis and I will explain our current findings.
Data e Ora : Friday, July 17, 2015 - 12:00
Relatore : Prof.ssa Nancy Amato
Affiliazione : Texas A&M University
Luogo : Sala riunioni DEI/G
Descrizione :
Abstract: ------------- We discuss new hybrid strategies we have developed for managing synchrony and memory in parallel graph algorithms. We first describe a new algorithmic paradigm k-level asynchronous (KLA) that enables the level of asynchrony in parallel graph algorithms to be parametrically varied from none (level-synchronous) to full (asynchronous). Results of an implementation of KLA in the STAPL Graph Library show excellent scalability on up to 96K cores and improvements of 10x or more over level-synchronous and asynchronous versions for graph algorithms such as breadth first search, PageRank, k-core decomposition and others on certain classes of real-world graphs. We next describe a novel RAM-Disk hybrid approach to graph processing that works by partitioning the graph into subgraphs that fit in RAM and uses a paging-like technique to load subgraphs. An implementation of this strategy in STAPL shows that without modifying the algorithms, this approach can scale from small memory-constrained systems (such as tablets) to largescale distributed machines with 16,000+ cores. Bio: ------------- Nancy M. Amato is Unocal Professor of Computer Science and Engineering at Texas A&M University where she co-directs the Parasol Lab. Her main areas of research focus are motion planning and robotics, computational biology and geometry, and parallel and distributed computing. Amato received undergraduate degrees in Mathematical Sciences and Economics from Stanford University, and M.S. and Ph.D. degrees in Computer Science from UC Berkeley and the University of Illinois, respectively. She was an AT&T Bell Laboratories PhD Scholar, received an NSF CAREER Award, is an ACM Distinguished Speaker, and was an IEEE Robotics and Automation Society Distinguished Lecturer (2006-2007). She is a AAAS Fellow and an IEEE Fellow.
Data e Ora : Friday, June 5, 2015 - 15:00
Relatore : Prof. Pierluigi Crescenzi
Affiliazione : Universita' di Firenze
Luogo : Sala riunioni DEI/G
Descrizione :
Abstract: Closeness is an important centrality measure widely used in the analysis of real-world complex networks. In particular, the problem of selecting the k most central nodes with respect to this measure has been deeply analyzed in the last decade. However, even for not very large networks, this problem is computationally intractable in practice: indeed, Abboud et al have recently shown that its complexity is strictly related to the complexity of the All-Pairs Shortest Path (in short, APSP) problem, for which no subcubic ``combinatorial'' algorithm is known. In this paper, we propose a new algorithm for selecting the k most closeness central nodes in a graph. In practice, this algorithm significantly improves over the APSP approach, even though its worst-case time complexity is the same. For example, the algorithm is able to compute the top $k$ nodes in few dozens of seconds even when applied to real-world networks with millions of nodes and edges. We will also experimentally prove that our algorithm drastically outperforms the most recently designed algorithm, proposed by Olsen et al. Finally, we apply the new algorithm to the computation of the most central actors in the IMDB collaboration network, where two actors are linked if they played together in a movie.Joint work with Michele Borassi and Andrea Marino.
Data e Ora : Friday, April 24, 2015 - 11:00
Relatore : Luigi Palopoli
Affiliazione : University of Trento
Luogo : Aula Magna "A. Lepschy"
Descrizione :
Modern embedded controllers are often characterised by an important level of variability in the computation requirements for the extraction of sensor data. What is more, obviouscost reasons demand sharing hardware between different concurrent activities.These facts strain the assumptions that underlie classical flows for the designof digital controllers: constant delays and infinite availability ofcommunication and computation resources.In this setting a system design with guaranteed control performance requires:
In this talk, we will show our results on each of these areas. The techniques we propose allows tackling such questions as: 1. the choice of the platformparameters that minimise the resource consumption of a control application with guaranteed levels for the Quality of Control, 2. find the best selection of theplatform parameters for a set of control application competing for the same platform.
Data e Ora : Monday, April 20, 2015 - 12:00
Relatore : Keshav Pingali
Affiliazione : The University of Texas at Austin
Luogo : Sala riunioni 318 DEI/G
Descrizione :
Task graphs or dependence graphs are used in runtime systems toschedule tasks for parallel execution. In problem domains such asdense linear algebra and signal processing, dependence graphs can begenerated from a program by static analysis. However, in problemdomains such as discrete-event simulation and simulation usingasynchronous variational integrators, the set of tasks and thedependences between tasks in a program are complex functions ofruntime values and cannot be determined statically.In this talk, we introduce a novel approach for exploiting parallelismin such programs. This approach is based on a data structure calledthe kinetic dependence graph (KDG), which consists of a dependencegraph together with update rules that incrementally update the graphto reflect changes in the dependence structure whenever a task iscompleted. We have implemented a simple programming model that allowsprogrammers to write these applications at a high level ofabstraction, and a runtime within the Galois system that builds theKDG automatically and executes the program in parallel. On a suite ofprograms that are difficult to parallelize otherwise, we have obtainedspeedups of up to 33 on 40 cores, out-performing third-partyimplementations in some cases.This is joint work with Amber Hassaan and Donald Nguyen at UT Austin.
Data e Ora : Friday, March 27, 2015 - 11:00
Relatore : Prof. Lorenzo Rosasco
Affiliazione : Universita' di Genova, MIT
Luogo : Aula Magna "A. Lepschy"
Descrizione :
Abstract: Availability of large high-dimensional data-sets has urged the development of optimization solutions for large scale learning problems. From a theoretical perspective this has motivated the goal of better understanding the interplay between statistics and optimization, towards developing new, more efficient learning algorithms. Indeed, while much theoretical work has been devoted to study statistical properties of estimators defined by variational schemes (a.k.a. Tikhonov regularization), and the computational properties of optimization procedures to solve the corresponding minimization problems, much less work has been devoted to the integration of statistical and optimization aspects. In this talk, we will present some recent proposals to develop machine learning algorithms which are provably efficient as well as statistically sound. In particular, we will discuss different instances of iterative regularization methods and, if time permit, randomized sampling techniques allowing further improvements. Short BIO: http://web.mit.edu/lrosasco/www/
Data e Ora : Thursday, January 22, 2015 - 16:30
Relatore : Prof. Umberto Vincenti
Affiliazione : Universita' di Padova
Luogo : Aula Magna 'A. Lepschy'
Descrizione :
Data e Ora : Thursday, January 8, 2015 - 17:00
Relatore : Prof. Umberto Vincenti
Affiliazione : Universita' di Padova
Luogo : Aula Magna 'A. Lepschy'
Descrizione :
