Computer engineering - Ingegneria informatica
The Engineering Department of the University of Perugia is one of the units involved in the PRIN 2012 project "AMANDA: Algorithms for MAssive and Networked DAta". The unit is coordinated by Prof. Giuseppe Liotta, and includes the researchers of the Computer Engineering group of the Department, namely Prof. Walter Didimo, Dr. Emilio Di Giacomo, Dr. Carla Binucci, Dr. Luca Grilli, and Dr. Fabrizio Montecchiani.
AMANDA will investigate algorithmics for massive data sets. On one hand the project will study emerging and realistic computational models and general algorithm design techniques; on the other hand it will focus on algorithmic issues specific for networked data sets. Pursuing these objectives raises hard research challenges, since the size of the data as well as their networked and evolving nature require a quantum leap in algorithmic design and engineering. These challenges are addressed in two workparts (WPs), each combining theoretical analysis with extensive experimental validation:
- WP1 Massive Data Sets - focused on a number of methodological issues that arise when processing very large datasets and on the design of novel algorithmic solutions for specific data-intensive applications.
- WP2 Massive and Evolving Networked Data Sets - focused on computing structural properties of massive and evolving networks and on designing ad-hoc algorithms and visual tools for supporting the mining process.
The presented research is the result of a collaboration among our research group, the Brain Research Center of the National Tsing Hua University of Taiwan and the National Center for High-performance Computing of Taiwan. Recently, the Brain Research Center of the National Tsing Hua University of Taiwan published a 3D image database for single neurons, called FlyCircuit (www.flycircuit.tw). This data source has been built from the brain of Drosophila Melanogaster and it contains more than ten thousand neurons. Single neuron images were acquired and reconstructed into a standardized brain space so as to build a 3D connection atlas with physiological significance. Furthermore, the connectivity matrix raising from this neural network has been constructed. The main goal of our collaboration was to analyze this reconstructed neural network, exploiting our different expertise in biology, bioinformatics, information visualization and algorithm engineering. Indeed, this research required an interdisciplinary collaboration involving systematic image collection, coordinates unification, algorithms for graph analysis and visualization.
We demonstrate the resiliency of this reconstructed neural network under various errors and attacks. Most importantly, our results showed that the robustness of the network improved throughout the network’s development. We ran several experiments to measure the network's robustness and we adopted a visualization algorithm for clustered networks to compare the network structure before and during an attack. The information provided by the drawings of the network gave us a fundamental support in designing the experiments and analyzing the results.
Our findings can provide new clues for successful network construction and shed new light on the generation process of this neural network. Being able to understand and replicate the conditions behind such a robust structure would have a great impact in designing reliable networks in different application domains.
- Hsiu-Ming Chang, Ann-Shyn Chiang, Walter Didimo, Ching-Yao Lin, Giuseppe Liotta and Fabrizio Montecchiani: On the Robustness of the Drosophila Neural Network. IEEE NSW 2013: 168-171