Belaid Moa

Termination Criteria in the Moore-Skelboe Algorithm for Global Optimization by Interval Arithmetic

We investigate unconstrained optimization with an objective function that has an unknown and possibly large number of local minima. By varying the selection and termination criteria, we obtain several variants of the Moore-Skelboe algorithm for distinct tasks in nonconvex global optimization. All of these terminate after having found the best answer that is possible, given the precision of the underlying hardware and given the expression for the objective function. The first algorithm finds the best lower bound for the global minimum. This is then extended to a version that adds an upper bound.

Design and Construction of a Big Data Analytics Framework for Health Applications

We propose to establish a framework for supporting Big Data Analytics (BDA) on real healthcare big data. To test the analytic framework, we used UVic WestGrid (4412 cores computer cluster) to analyze the emulation of 10 billion healthcare records that represented the main hospital system and its reporting via its data warehouse stored at Vancouver Island Health Authority (VIHA). The study showed that the build of the BDA platform requires changes to the configurations to the MapReduce component of Hadoop (HDFS) and to the indexing of HBASE. The ingestion and replication of the data over a large volume iteratively offers a method for data migration of large volumes of real healthcare data via HDFS and to query in that some distributed filing system. Furthermore, the query performance was very satisfied via Apache Phoenix layer that is run in parallel across all nodes on HBASE. The study has demonstrated that the proposed BDA process and configuration met patient data security and performance requirements of healthcare BDA.

Unsupervised nonparametric classification of polarimetric SAR data using the K-nearest neighbor graph

Polarimetric SAR classifications are often based on assumptions about the shape of clusters in the data space. Such a scheme will fail for nonlinear structures in the feature space, unless the classification algorithm has the capacity to describe cluster shapes in sufficient generality. Existing polarimetric SAR classification methods are faced by this exact problem: typically they initialize clusters in the Cloude-Pottier parameter space [1], further optimizing them in the coherency matrix space [2, 3]. Methods using K-means [2] or agglomeration [3] require clusters that are spherical, or compact and well separated, respectively. In the Cloude-Pottier space, these requirements are not met, so initialization in the Cloude-Pottier space cannot be consistent with optimization by K-means or agglomeration. This paper sets out to address this problem, by implementing a new data-driven clustering approach, for arbitrarily shaped clusters. It is applied to quad-polarisation data, demonstrating the new methodologys potential for forest land-cover type discrimination.

Magnetic Anisotropy and Magnetoresistance Properties of Co/Au Multilayers

We report on the magnetic anisotropy (MA) and magnetoresistance (MR) properties of [Co (t_Co)/Au (t_Au)]_N^θ multilayers prepared by changing the incident angle of deposition, θ, thicknesses t_Co and t_Au of Co and Au layers, respectively, and number of bilayers, N, while keeping N × t_Co fixed at 20 nm. Although all of the multilayers showed MAs at low applied magnetic fields, H, and room temperature, the ones deposited at θ = 45° showed remarkable magnetic anisotropies, which is further enhanced upon magnetic annealing. The [Co (1 nm)/Au (2 nm)]₂₀⁴⁵ multilayer showed a maximum MR of 2.1% at room temperature and H = 1 kOe. Moreover, for the same field H and current I, the transverse (H ⊥ I) MR is always larger than the longitudinal (H ∥ I) MR.

Interactive Healthcare Big Data Analytics Platform under Simulated Performance

To utilize data from hospital systems, big data analytics (BDA) has become increasingly important. BDA enable queries of large highly diverse and real volumes of patient data in an interactively dynamic way that enriches the use of the platform with data visualization for healthcare. We established a Healthcare BDA (HBDA) platform at the University of Victoria (UVic) with Compute Canada/Westgrid, and Vancouver Island Health Authority (VIHA), Victoria, BC, Canada. The framework was a proof-of-concept implementation that tested emulated patient data representative of the main hospital system at VIHA. We cross-referenced all data, its profiles and metadata, with the existing clinical reporting. Our HBDA platform and its performance was tested for different patient query types in simulation with the data ingested into Hadoop file system over different applications of Apache Spark with Zeppelin and Jupyter web-based interfaces, and Apache Drill interfaces. The results showed that the ingestion time of one billion records took circa 2 hours via Apache Spark. Apache Drill outperformed Spark/Zeppelin and Spark/Jupyter. However, it was restricted to running more simplified queries, and very limited in its visualizations exhibiting poor usability for healthcare. Zeppelin running on Spark showed ease-of-use interactions for health applications, but it lacked the flexibility of its interface tools and required extra setup time before running queries. Jupyter on Spark offered high performance stacks not only over our HBDA platform but also in unison to run all queries simultaneously with high usability for a variety of reporting requirements by providers and health professionals.

High Performance Proactive Digital Forensics

With the increase in the number of digital crimes and in their sophistication, High Performance Computing (HPC) is becoming a must in Digital Forensics (DF). According to the FBI annual report, the size of data processed during the 2010 fiscal year reached 3,086 TB (compared to 2,334 TB in 2009) and the number of agencies that requested Regional Computer Forensics Laboratory assistance increasing from 689 in 2009 to 722 in 2010. Since most investigation tools are both I/O and CPU bound, the next-generation DF tools are required to be distributed and offer HPC capabilities. The need for HPC is even more evident in investigating crimes on clouds or when proactive DF analysis and on-site investigation, requiring semi-real time processing, are performed. Although overcoming the performance challenge is a major goal in DF, as far as we know, there is almost no research on HPC-DF except for few papers. As such, in this work, we extend our work on the need of a proactive system and present a high performance automated proactive digital forensic system. The most expensive phase of the system, namely proactive analysis and detection, uses a parallel extension of the iterative z algorithm. It also implements new parallel information-based outlier detection algorithms to proactively and forensically handle suspicious activities. To analyse a large number of targets and events and continuously do so (to capture the dynamics of the system), we rely on a multi-resolution approach to explore the digital forensic space. Data set from the Honeynet Forensic Challenge in 2001 is used to evaluate the system from DF and HPC perspectives.

Tabular Expressions Operators

Tabular expressions are one of the most important table-based techniques used to formally specify software requirements. The power of tabular expressions stems from their visual structure, and concise representation of mathematical functions and relations. Towards using tabular expressions as a programming language, we propose a tabular expression language in which tabular expressions are first class citizens. The language is built upon atomic tabular expressions and operators. As such, tabular expressions are viewed as a stack of atomic expressions and operators that we apply on them. This view enhances building tools supporting the semantics of tabular expressions, and using them directly during the implementation. The tabular operators introduced are used to compose and decompose tabular expressions in a modular way, which improves their semantics.

A framework for efficiently parallelizing nonlinear noise reduction algorithm

In hyperspectral imagery, noise reduction is a vital and common pre-processing step that needs to be executed accurately and efficiently. Until recently, hyperspectral data was modeled using linear stochastic processes and the noise was assumed to manifest itself in a narrow spatial frequency band. The signal and noise are thus considered independent and most of the proposed noise reduction algorithms transform the hyperspectral data linearly from one space to another for noise and signal separation. Hyperspectral data, however, exhibits nonlinear characteristics making the noise frequency and signal dependent [1, 2]. Therefore, to accurately reduce the noise in hyperspectral data, a nonlinear noise reduction algorithm, such as the one we propose in this paper, must be considered. The algorithm, however, is computationally expensive and requires parallelization. To this end, we offer a framework which we have implemented and evaluated.

Data Sources and Datasets for Cloud Intrusion Detection Modeling and Evaluation

Over the past few years cloud computing has skyrocketed in popularity within the IT industry. Shifting towards cloud computing is attracting not only industry but also government and academia. However, given their stringent privacy and security policies, this shift is still hindered by many security concerns related to the cloud computing features, namely shared resources, virtualization and multi-tenancy. These security concerns vary from privacy threats and lack of transparency to intrusions from within and outside the cloud infrastructure. Therefore, to overcome these concerns and establish a strong trust in cloud computing, there is a need to develop adequate security mechanisms for effectively handling the threats faced in the cloud. Intrusion Detection Systems (IDSs) represent an important part of such mechanisms. Developing cloud based IDS that can capture suspicious activity or threats, and prevent attacks and data leakage from both inside and outside the cloud environment is paramount. One of the most significant hurdles for developing such cloud IDS is the lack of publicly available datasets collected from a real cloud computing environment. In this chapter, we discuss specific requirements and characteristics of cloud IDS in the light of traditional IDS. We then introduce the first public dataset of its kind for cloud intrusion detection. The dataset consists of several terabytes of data, involving normal activities and multiple attack scenarios, collected over multiple periods of time in a real cloud environment. This is an important step for the industry and academia towards developing and evaluating realistic intrusion models for cloud computing.

Simulations of Hadoop/MapReduce-Based Platform to Support its Usability of Big Data Analytics in Healthcare

In many hospital systems, new technologies that influence patient data require extensive technical testing before implementation into production. Therefore, to implement, an existing High Performance Computing (HPC) Linux node clusters via WestGrid were used to represent a simulation of patient data benchmarked and cross-referenced with current metadata profiles in operational hospital systems at the Vancouver Island Health Authority (VIHA), Victoria, Canada. Over the tested cross-platform, the data were generated, indexed and stored over a Hadoop Distributed File System (HDFS) to noSQL database (HBase) that represented three billion patient records. The study objective to establish an interactive Big Data Platform (BDA) was successful implemented in that Hadoop/MapReduce technologies formed the framework of the platform distributed with HBase (key-value NoSQL database storage) and generated desired hospital-specific metadata at very extremely large volumes. In fact, the framework over generated HBase data files took a week or a month for one billion (10TB) and three billion (30TB), respectively. Further performance tests retrieved results from simulated patient records with Apache tools in Hadoop’s ecosystem. At optimized iteration, HDFS ingestion with HBase exhibited sustained database integrity over hundreds of iterations; however, to complete the bulk loading via MapReduce to HBase required a month. Inconsistencies of MapReduce limited the capacity to generate/replicate data to HBase efficiently. Hospital system based on patient encounter database was very difficult and data profiles were fully representative of complex patient-tohospital relationships. Our platform is important to lead discovery of useful big data technologies across multiple hospital systems.