When : Thursday, December 20, 2007 - 15:00
Speaker : Ing. Giampaolo Tomasi
Affiliation : DEI
Where : Aula Magna `A. Lepschy`
Description :

The generation of parametric maps -i.e. maps that associate to each pixel of the image its kinetic parameters- is a crucial issue in Positron Emission Tomography. Unfortunately, the quality of such maps is sometimes poor, due to the high noise of data at pixel level. In this talk, different types of Bayesian methods, which improve the quality and reliability of parametric maps, will be described and their potentialities showed both on simulated and real data sets.

When : Monday, December 17, 2007 - 15:00
Speaker : Prof. Alessandro Panconesi
Affiliation : Universit� di Roma La Sapienza
Where : Aula Magna `A. Lepschy`
Description :

Given a set of points S and a query point q, the k-nearest neighbour (k-NN) problem is to find the k points in S that are closest to q. This geometric problem is ubiquitous in (web) information retrieval and efficient algorithms are hard to come by, in spite of much work done in the area. In this talk we will introduce a very simple random clustering (RC) technique and show that: (1) RC can be analyzed mathematically and strong, useful properties can be proven about it; (2) with real data, RC outperforms well-known solutions for k-NN such as p-spheres and rank aggregation. Our experiments show that RC is superior to these well-known methods for a variety of applications, including text retrieval and image processing. Our theorems explain why.
Joint work with: Flavio Chierichetti, Prabakhar Raghavan, Mauro Sozio, Alessandro Tiberi, and Eli Upfal

When : Thursday, December 6, 2007 - 15:00
Speaker : Ing. Marco Falda
Affiliation : DEI
Where : Aula Magna `A. Lepschy`
Description :

Many applications developed in very different areas such as Planning, Natural Language Processing, Hardware Design and Bioinformatics, need to manage critical aspects involving time. The models investigated in Artificial Intelligence literature for Temporal Reasoning are mainly based on the Constraint Satisfaction Problem (CSP) paradigm. Some of them deal with the representation of quantitative aspects of temporal data in terms of metric temporal statements about points. Other approaches, such as Interval Algebra, deal with a qualitative representation of temporal information in terms of qualitative relationships between intervals.
In everyday life these two aspects (either quantitative or qualitative) of temporal knowledge are not distinguished and real data are often affected by imprecision and uncertainty. Therefore, it is important to establish expressive frameworks for representing complex temporal relationships in a unified way.
In this talk the problem of representing and reasoning with fuzzy temporal knowledge in a general and flexible manner is introduced and a model for integrating qualitative and quantitative fuzzy temporal constraints is discussed.

When : Thursday, November 22, 2007 - 15:00
Speaker : Ing. Elena Fasolo
Affiliation : DEI
Where : Aula Magna `A. Lepschy`
Description :

Data Dissemination consists on spreading a large amount of information to all nodes belonging to a network. Thus it can be applied for different purposes in a lot of practical scenarios. The peculiar characteristics of the system in use make the definition of efficient data dissemination schemes an interesting and challenging goal. In particular, developing efficient algorithms for wireless ad hoc networks is still an open issues due to the broadcast nature of the channel and to the need of managing all data transmissions in a distributed way. The former leads to a lot of problems related to the channel contention, collisions and interference. The latter requires to define algorithms which exploit only local information of the network and which are scalable and robust to the node mobility.
In this presentation we investigate how data dissemination schemes can be enhanced by the use of network coding. Network coding is a recent paradigm applied at network layer to increase the throughput. A lot of studies showed its effectiveness form a theoretical point of view. We, instead, investigate practical aspects related to the implementation of data dissemination schemes based on network coding in realistic environments affected by interference, collisions, fading and so on.

When : Thursday, November 15, 2007 - 15:00
Speaker : Ing. Ottavio Campana
Affiliation : DEI
Where : Aula Magna `A. Lepschy`
Description :

Transmission of encoded video sequences over unreliable networks usually requires the adoption of protection techniques to guarantee good reconstruction quality at the receiver side. Multiple Description Coding (MDC) strategies add reliability to real-time video applications, where retransmission is not possible and packet losses afflict several frames degrading the overall quality.
In Multiple Description Coding, the transmitter splits the original signal into several correlated streams, called descriptions, which are sent to the receiver through distinct channels. Whenever all descriptions are correctly received, the signal is completely decoded at full coding quality, while in case of partial loss of some descriptions, lost information is estimated exploiting the correlation between descriptions and a coarse-quality signal is reconstructed.
In this talk, after a brief introduction to video and multiple description coding, applications of MDC to the H.264/AVC standard and Scalable Video will be presented.

When : Thursday, November 8, 2007 - 15:00
Speaker : Dr. Alessio Botta
Affiliation : IMT Lucca Institute for Advanced Studies, Lucca, Italy
Where : Aula Magna `A. Lepschy`
Description :

Evolutionary algorithms provide an off-the-shelf set of meta-heuristic techniques that can be used to solve complex, constrained and non-linear problems, including traditional optimization ones, such as models for scheduling and routing in wireless networks. In this colloquium, we will show technical details on how a basic evolutionary algorithm can be tailored to suit network optimization problems. Further, we will quickly review advanced evolutionary algorithms topics, such as co-evolution, constraint handling and multi-objective optimization, and provide some hints on how these techniques may be used to perform scalable, distributed and robust network problem solving.