Title: Tolerating uncertainty via evolvable-by-design software
Uncertainty is ubiquitous when software is designed. Requirements are often uncertain, and volatile. Assumptions about the behavior of the environment in which the software will be embedded are also often uncertain. The virtual platform on which the software will be operated may likewise be subject to uncertain operating conditions. Design-time uncertainty is resolved during operation, and often the way it is resolved changes over time. This leads to the need for software to evolve continuously, to keep guaranteeing satisfaction of its quality goals.
Evolution can partly be self-managed, by adding self-adaptive capabilities to the software. This requires an upfront careful analysis to understand where the sources of uncertainty, how they can be resolved during operation, and how they can be managed through dynamic reconfigurations. Whenever self-adaptation cannot solve the problems, designers must be in the loop to provide new solutions that can be dynamically incorporated in the running system.
The talk provides a holistic view of how to handle uncertainty, which is based on the notion of perpetual development and adaptation. It shows that existing approaches to software development need to be rethought to respond to these challenges. The traditional separation between development and operation (design time and run time) blurs and even fades. The talk especially focuses on modeling and verification, which need to be rethought in the light of perpetual development and evolution. It also focuses on achieving self-adaptation to support continuous satisfaction of non-functional requirements --- such as reliability, performance, energy consumption --- in the context of virtualized environments (cloud computing, service-oriented computing).
Carlo Ghezzi is an ACM Fellow (1999), an IEEE Fellow (2005), a member of the European Academy of Sciences and of the Italian Academy of Sciences. He received the ACM SIGSOFT Outstanding Research Award (2015) and the Distinguished Service Award (2006). He is the current President of Informatics Europe.
He is a regular member of the program committee of flagship conferences in the software engineering field, such as the ICSE and ESEC/FSE, for which he also served as Program and General Chair.
He has been the Editor in Chief of the ACM Trans. on Software Engineering and Methodology and an associate editor of and IEEE Trans. on Software Engineering. Currently he is an Associate Editor of the Communications of the ACM and Science of Computer Programming.
Ghezzi’s research has been mostly focusing on different aspects of software engineering. He co-authored over 200 papers and 8 books. He coordinated several national and international research projects. He has been the recipient of an ERC Advanced Grant.
Title: From Ensemble Learning to Learning in the Model Space
Ensemble learning has been shown to be very effective in solving many
challenging regression and classification problems. Multi-objective learning
offers not only a novel method to construct and learn ensembles automatically,
but also better ways to balance accuracy and diversity in an ensemble. This
talk introduces the basic ideas behind multi-objective learning. It describes
how ensembles can be used in mining data streams from the point of view of
online learning. In particular, the importance of diversity in online learning
is demonstrated. Finally, a novel approach to data stream mining is presented
--- learning in the model space, which can handle very challenging data
streams. The effectiveness of such an approach is illustrated by concrete
examples in cognitie fault diagnosis.
Xin Yao is a Chair Professor of Computer Science at the Southern University of
Science and Technology in Shenzhen, China. He is an IEEE Fellow and
the President (2014-15) of IEEE Computational Intelligence Society (CIS).
His work won the 2001 IEEE Donald G. Fink Prize Paper Award, 2010 IEEE
Transactions on Evolutionary Computation Outstanding Paper Award, 2010 BT
Gordon Radley Award for Best Author of Innovation (Finalist), 2011 and 2015
IEEE Transactions on Neural Networks Outstanding Paper Awards, and many other
best paper awards. He won the prestigious Royal Society Wolfson Research
Merit Award in 2012 and the IEEE CIS Evolutionary Computation Pioneer Award
in 2013. He was the Editor-in-Chief (2003-08) of IEEE Transactions on
Evolutionary Computation and is an Associate Editor or Editorial Member of
more than ten other journals. His major research interests include
evolutionary computation, ensemble learning, and their applications,
especially in software engineering.
Title: Human Millimeter-wave Holographic Imaging and Automatic Target Recognition
Millimeter-wave (MMW) holographic Imaging is one of the most effective methods for human inspection because it can acquire three-dimensional images of
human body via a single scan. Due to its high penetration through fabrics and contrast in reflectivity, we can easily distinguish contrabands such as guns
and explosives on human body in MMW images. Besides, millimeter wave is non-ionizing radiation of no potential health threat. Our imaging system utilizes a
linear antenna array to improve scanning speed. The image reconstruction is achieved via Fast Fourier Transform (FFT) and spatial spherical wave expansion.
Linear antenna array will results in artifacts in the reconstructed images, system errors and background scattering could also have negative influences on
MMW images. We propose a set of calibration and denoising methods to eliminate these influences and obtain significant image quality in experimental studies.
Our experiments indicate that these methods could improve the image quality. Automatic Target Recognition (ATR) based on millimeter-wave holographic Images
is a key step to meet the requirements of intelligent devices. Object detection methods for color images are not very efficient in human body MMW images.
Thus, we proposed a synthetic object detection method for MMW images on the basis of machine learning. According to previous theories, both multi-layer model
and sparse coding could improve the accuracy of recognition. Thus, we select saliency, SIFT and HOG features to describe MMW images and build a two-layer
model to encode these features. The encoded features are fed to a linear SVM for target/non-target classification. As the amount of training data contributes
to the efficiency of SVM classifier, we build a training set consists of over 30,000 human body MMW images which is generated from the 3,154 original images
via several image augmentation techniques. The experimental results indicate that the total target detection rate in MMW images is improved from 70% to 85% by
training set augmentation, which demonstrate the efficiency of our method.
Dr. Zhao Ziran received his B.S. and Ph.D. from the Tsinghua University, in 1998 and 2004, respectively. In 1994, he joined the Tsinghua University and
became associate professor in 2008. He received a joint appointment as executive deputy director of Institute for Security Detection Technology in 2012.
Dr. Zhao’s research interest is generally in the area of imaging and detection technology. In particular, he works to apply new image reconstruction
algorithms to a wide range of applications including millimeter-wave imaging, terahertz imaging, radiation imaging, cosmic-ray muon tomography. He has being
devoted in solving scientific and technical problems on security detection technology and providing hi-tech equipment for anti-terrorism. He won the National
Patent Golden award of 2009. And he is the main member of Tsinghua university radiation imaging innovative research team, which won the National Science and
Technology Progress Award (Innovative Research Team) in 2013.