We perform a detailed analysis of the C++ implementation of the Cluster Affiliation Model for Big Networks (BigClam) on the Stanford Network Analysis Project (SNAP). BigClam is a popular graph mining algorithm that is capable of finding overlapping communities in networks containing millions of nodes. Our analysis shows a key stage of the algorithm – determining if a node belongs to a community – dominates the runtime of the implementation, yet the computation is not parallelized. We show that by parallelizing computations across multiple threads using OpenMP we can speed up the algorithm by 5.3 times when solving large networks for communities, while preserving the integrity of the program and the result. 2012 ACM Subject Classification Computing methodologies → Concurrent algorithms
One of the challenges our society faces is the ever increasing amount of data. Among existing platforms that address the system requirements, Hadoop is a framework widely used to store and analyze “big data”. On the human side, one of the aids to finding the things people really want is recommendation systems. This paper evaluates highly scalable parallel algorithms for recommendation systems with application to very large data sets. A particular goal is to evaluate an open source Java message passing library for parallel computing called MPJ Express, which has been integrated with Hadoop. As a demonstration we use MPJ Express to implement collaborative filtering on various data sets using the algorithm ALSWR (Alternating-Least-Squares with Weighted-λ-Regularization). We benchmark the performance and demonstrate parallel speedup on Movielens and Yahoo Music data sets, comparing our results with two other frameworks: Mahout and Spark. Our results indicate that MPJ Express implementation of ALSWR has very competitive performance and scalability in comparison with the two other frameworks. 2012 ACM Subject Classification Information systems→ Database management system engines
Gesture Recognition and Classification using Intelligent Systems Norah Alnaim and Maysam Abbod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8:1–8:1 KubeNow: A Cloud Agnostic Platform for Microservice-Oriented Applications Marco Capuccini, Anders Larsson, Salman Toor, and Ola Spjuth . . . . . . . . . . . . . . . . . 9:1–9:2 ICCSW17 2017 Imperial College Computing Student Workshop (ICCSW 2017). Editors: Fergus Leahy and Juliana Franco Open Access Series in Informatics Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, Germany
Traditional biometric systems are designed and configured to operate in predefined circumstances to address the needs of a particular application. The performance of such biometrics systems tend to decrease because when they encounter varying conditions as they are unable to adapt to such variations. Many real-life scenarios require identification systems to recognise uncooperative people in uncontrolled environments. Therefore, there is a real need to design biometric systems that are aware of their context and be able to adapt to changing conditions. The context-awareness and adaptation of a biometric system are based on a set of factors that include: the application (e.g. healthcare system, border control, unlock smart devices), environment (e.g. quiet/noisy, indoor/outdoor), desired and pre-defined requirements (e.g. speed, usability, reliability, accuracy, robustness to high/low quality samples), user of the system (e.g. cooperative or non-cooperative), the chosen modality (e.g. face, speech, gesture signature), and used techniques (e.g. pre-processing to normalise and clean biometrics data, feature extraction and classification). These factors are linked and might affect each other, hence the system has to work adaptively to meet its overall aim based to its operational context. The aim of this research is to develop a multiagent based framework to represent a contextaware adaptive biometric system. This is to improve the decision making process at each processing step of traditional biometric identification systems. Agents will be used to provide the system with intelligence, adaptation, flexibility, automation, and reliability during the identification process. The framework will accommodate at least five agents, one for each of the five main processing steps of a typical biometric system (i.e. data capture, pre-processing, feature extraction, classification and decision). Each agent can contribute differently towards its designated goal to achieve the best possible solution by selecting/ applying the best technique. For example, an agent can be used to assess the quality of the input biometric sample to ensure the important features can be extracted and processed in further steps. Another agent can be used to pre-process the biometric sample if necessary. A third agent is used to select the appropriate set of features followed by another to select a suitable classifier that works well in a given condition. 2012 ACM Subject Classification Security and privacy → Biometrics
We perform a detailed analysis of the C++ implementation of the Cluster Affiliation Model for Big Networks (BigClam) on the Stanford Network Analysis Project (SNAP). BigClam is a popular graph mining algorithm that is capable of finding overlapping communities in networks containing millions of nodes. Our analysis shows a key stage of the algorithm - determining if a node belongs to a community - dominates the runtime of the implementation, yet the computation is not parallelized. We show that by parallelizing computations across multiple threads using OpenMP we can speed up the algorithm by 5.3 times when solving large networks for communities, while preserving the integrity of the program and the result.
Artificial Intelligence is increasingly being used to both augment existing fields of research and open up new avenues of discovery. From quality control for imaging flow cytometry to computational musicology, modern AI is an exciting new tool for research and thus knowing how to engineer AI systems in a research context is a vital new skill for RSEs to acquire. In this talk, I will outline four different areas of AI: supervised learning, unsupervised learning, interactive learning, and Bayesian learning. For each of these approaches, I will discuss how they typically map to different research problems and explore best practices for RSEs via specific use cases. At the end of the talk, you will have received a high-level overview of AI technologies and their use in research, have seen some cool examples of how AI has been used in a wide range of research areas, and have a good sense of where to go to learn more. 2012 ACM Subject Classification Computing methodologies → Artificial intelligence
Information retrieval (IR) has become one of the most popular Natural Language Processing (NLP) applications. Part of speech (PoS) parsing and tagging plays an important role in IR systems. A broad range of PoS parsers and taggers tools have been proposed with the aim of helping to find a solution for the information retrieval problems, but most of these are tools based on generic NLP tags which do not capture domain-related information. In this research, we present a domain-specific parsing and tagging approach that uses not only generic PoS tags but also domain-specific PoS tags, grammatical rules, and domain knowledge. Experimental results show that our approach has a good level of accuracy when applying it to different domains.
This paper presents the first dataset for eye segmentation in low resolution images. Although eye segmentation has long been a vital preprocessing step in biometric applications, this work is the first to focus on low resolutions image that can be expected from a consumer-grade camera under conventional human-computer interaction and / or video-chat scenarios. Existing eye datasets have multiple limitations, including: (a) datasets only contain high resolution images; (b) datasets did not include enough pose variations; (c) a utility landmark ground truth did not be provided; (d) high accurate pixel-level ground truths had not be given. Our dataset meets all the above conditions and requirements for different segmentation methods. Besides, a baseline experiment has been performed on our dataset to evaluate the performances of landmark models (Active Appearance Model, Ensemble Regression Tree and Supervised Descent Method) and deep semantic segmentation models (Atrous convolutional neural network with conditional random field). Since the novelty of our dataset is to segment the iris and the sclera areas, we evaluate above models on sclera and iris only respectively in order to indicate the feasibility on eye-partial segmentation tasks. In conclusion, based on our dataset, deep segmentation methods performed better in terms of IOU-based ROC curves and it showed potential abilities on low-resolution eye segmentation task.
Feature selection is typically employed before or in conjunction with classification algorithms to reduce the feature dimensionality and improve the classification performance, as well as reduce processing time. While particular approaches have been developed for feature selection, such as filter and wrapper approaches, some algorithms perform feature selection through their learning strategy. In this paper, we are investigating the effect of the implicit feature selection of the PRISM algorithm, which is rule-based, when compared with the wrapper feature selection approach employing four popular algorithms: decision trees, naive bayes, k-nearest neighbors and support vector machine. Moreover, we investigate the performance of the algorithms on target classes, i.e. where the aim is to identify one or more phenomena and distinguish them from their absence (i.e. non-target classes), such as when identifying benign and malign cancer (two target classes) vs. non-cancer (the non-target class).
Reliable event detection is one of the hottest research areas in the wireless sensor networks field these days. Battlefield monitoring, fire detection, nuclear and chemical attack, and gas leak detection are examples of the event detection applications. One of the main goals to WSNs is transmitting the sensed data to the sink (Base station) in an efficient way with minimum energy usage to achieve high degree of event detection reliability. Thus, Its very important to determine the reliability degree to know the number of data that are required to receive at the sink to achieve the desired reliability. Most of the previous research works proposed different solutions for reliable event detection. The idea of all these solutions is based on increasing the amount of the transmitted data to the sink by controlling the sources reporting rate. However, rising the reporting rate may lead to losing the transmitted data due to the network congestion and packets collision, and this is related to the restricted resources capacity of the network’s sensor nodes. Therefore, in this paper, a new indoor method to achieve quality based event reliability for critical event detection have been implemented using hardware sensor nodes (Waspmote). The idea of this method is depending on sending the sensed data to the sink using a node called Cluster Head (CH) in a sequence according to their priority from the high to the low. The network nodes have been deployed in the experiment area into clusters, and each cluster have a CH node which work on collecting the cluster members readings and reorder it in descending order to send it next to the sink. The probability to deliver the important data to detect the event to the sink will increase by using this new method. The proposed mechanism intends to improve the event detection reliability, minimize the end-to-end delay, and increase the network lifetime. Experiments results show that the proposed method achieved a good the performance in terms of packets delivery, event detection, and end-to-end delay. 2012 ACM Subject Classification Networks
Security policy and access control models are often based on qualitative attributes, e.g. security labels, cryptographic credentials. In this paper, we enrich one such model, namely the Chinese Walls model, with quantitative attributes derived from data. Therefore, we advocate a datadriven approach that considers a quantitative definition of access we term, working relations.
The main goal of this research is to contribute to automated performance anomaly detection for large-scale and complex distributed systems, especially for Big Data applications within cloud computing. The main points that we will investigate are: - Automated detection of anomalous performance behaviors by finding the relevant performance metrics with which to characterize behavior of systems. - Performance anomaly localization: To pinpoint the cause of a performance anomaly due to internal or external faults. - Investigation of the possibility of anomaly prediction. Failure prediction aims to determine the possible occurrences of catastrophic events in the near future and will enable system developers to utilize effective monitoring solutions to guarantee system availability. - Assessment for the potential of hybrid methods that combine machine learning with traditional methods used in performance for anomaly detection. The topic of this research proposal will offer me the opportunity to more deeply apply my interest in the field of performance anomaly detection and prediction by investigating and using novel optimization strategies. In addition, this research provides a very interesting case of utilizing the anomaly detection techniques in a large-scale Big Data and cloud computing environment. Among the various Big Data technologies, in-memory processing technology like Apache Spark has become widely adopted by industries as result of its speed, generality, ease of use, and compatibility with other Big Data systems. Although Spark is developing gradually, currently there are still shortages in comprehensive performance analyses that specifically build for Spark and are used to detect performance anomalies. Therefore, this raises my interest in addressing this challenge by investigating new hybrid learning techniques for anomaly detection in large-scale and complex systems, especially for in-memory processing Big Data platforms within cloud computing.
Writing papers is a core research skill for any researcher, but they aren't easy. Writing is not just a way to report on great research; it's a way to do great research. Yet many papers are so badly written that, even if they describe excellent work, the work has much less impact than it should. In this talk I'll give you seven simple, actionable guidelines that will, I hope, help you to write better papers, and have more fun at the same time. I don’t have all the answers—far from it—and I hope that the presentation will evolve into a discussion in which you share your own insights, rather than a lecture.
Gesture Recognition is defined as non-verbal human motions used as a method of communication in HCI interfaces. In a virtual reality system, gestures can be used to navigate, control, or interact with a computer. Having a person make gestures formed in specific ways to be detected by a device, like a camera, is the foundation of gesture recognition. Finger tracking is an interesting principle which deals with three primary parts of computer vision: segmentation of the finger, detection of finger parts, and tracking of the finger. Fingers are most commonly used in varying gesture recognition systems. Finger gestures can be detected using any type of camera; keeping in mind that different cameras will yield different resolution qualities. 2-dimensional cameras exhibit the ability to detect most finger motions in a constant surface called 2-D. While the image processes, the system prepares to receive the whole image so that it may be tracked using image processing tools. Artificial intelligence releases many classifiers, each one with the ability to classify data, that rely on its configuration and capabilities. In this work, the aim is to develop a system for finger motion acquisition in 2-D using feature extraction algorithms such as Wavelets transform (WL) and Empirical Mode Decomposition (EMD) plus Artificial Neural Network (ANN) classifier.
Linear discriminative models, in the form of Logistic Regression, are a popular choice within the clinical domain in the development of risk models. Logistic regression is commonly used as it offers explanatory information in addition to its predictive capabilities. In some examples the coefficients from these models have been used to determine overly simplified clinical risk scores. Such models are constrained to modeling linear relationships between the variables and the class despite it known that this relationship is not always linear. This paper compares the conditions under which linear discriminative and linear generative models perform best. This is done through comparing logistic regression and naïve Bayes on real clinical data. The work shows that generative models perform best when the internal representation of the data is closer to the true distribution of the data and when there is a very small difference between the means of the classes. When looking at variables such as sodium it is shown that logistic regression can not model the observed risk as it is non-linear in its nature, whereas naïve Bayes gives a better estimation of risk. The work concludes that the risk estimations derived from discriminative models such as logistic regression need to be considered in the wider context of the true risk observed within the dataset. 1998 ACM Subject Classification D.4.8 Performance
A whirlwind tour through the history and state-of-the-art of JavaScript execution and optimization, with a focus on the V8 engine used by Chrome and Node.js, and how a 10-day prototype became one of the most important programming languages in the world. Leszek Swirski has been a software engineer at Google for two years, first in California, now in London, working on the performance of the Android camera and the V8 Javascript engine. Before joining Google, Leszek did a PhD in the University of Cambridge, researching gaze estimation (a.k.a. eye tracking) on stereoscopic (a.k.a. “3D”) displays. 1998 ACM Subject Classification D.3.3 Language Constructs and Features, K.2 History of Computing
Dictionary Learning and Representation Learning are machine learning methods for decomposition, denoising and reconstruction of data with a wide range of applications such as text recognition, image processing and biological processes understanding. In this work we present DALILA, a scientific Python library for regularised dictionary learning and regularised representation learning that allows to impose prior knowledge, if available. DALILA, differently from the others available libraries for this purpose, is flexible and modular. DALILA is designed to be easily extended for custom needs. Moreover, it is compliant with the most widespread ML Python library and this allows for a straightforward usage and integration. We here present and discuss the theoretical aspects and discuss its strength points and implementation. 1998 ACM Subject Classification G.1.6 Optimization, D.1.3 Concurrent Programming, D.2.2 Design Tools and Techniques
KubeNow is a platform for rapid and continuous deployment of microservice-based applications over cloud infrastructure. Within the field of software engineering, the microservice-based architecture is a methodology in which complex applications are divided into smaller, more narrow services. These services are independently deployable and compatible with each other like building blocks. These blocks can be combined in multiple ways, according to specific use cases. Microservices are designed around a few concepts: they offer a minimal and complete set of features, they are portable and platform independent, they are accessible through language agnostic APIs and they are encouraged to use standard data formats. These characteristics promote separation of concerns, isolation and interoperability, while coupling nicely with test-driven development. Among many others, some well-known companies that build their software around microservices are: Google, Amazon, PayPal Holdings Inc. and Netflix [11].
ZooKeeper is a crash-tolerant system that offers fundamental services to Internet-scale applications, thereby reducing the development and hosting of the latter. It consists of N ≥ 3 servers that form a replicated state machine. Maintaining these replicas in a mutually consistent state requires executing an Atomic Broadcast Protocol, Zab, so that concurrent requests for state changes are serialised identically at all replicas before being acted upon. Thus, ZooKeeper performance for update operations is determined by Zab performance. We contribute by presenting two easy-to-implement Zab variants, called ZabAC and ZabAA. They are designed to offer small atomic-broadcast latencies and to reduce the processing load on the primary node that plays a leading role in Zab. The former improves ZooKeeper performance and the latter enables ZooKeeper to face more challenging load conditions. 1998 ACM Subject Classification D.2.8 Performance measures, D.4.7 Distributed systems
