Abstract: Online social networking is becoming an integral part of people’s daily life withthe penetration of mobile computing and communication devices. In this research,we propose a novel social networking mobile application for university studentsto augment their campus life using the short range radios on mobile devices toform an infrastructureless wireless network. Such a mobile application is aresearch prototype to study Delay-Tolerant Networking, a new computernetworking paradigm featuring “store-carry-forward”. Using this prototypeapplication, we collect information about how social networking messagespropagate through the network. By analyzing such data, we can devise moreeffective and efficient data transfer mechanisms in such networks.

* * *

0930 Mark Hatcher, Ph.D. Computer Science

Title: Modelling with Timed Automata

Abstract: Checking the properties (such as liveness and the absence of deadlocks) andevaluating the performance of complex, concurrent systems is becomingincreasingly difficult because of the convoluted interactions amongst thecomponents of such systems. A popular approach is to derive a behavioural modelof all components of the original system, and to study the properties of thismodel. Timed automata are often used as formal models of discrete-eventcomponents and their interactions. Systems of interacting automata can beevaluated by using structural methods or, more often, can provide performancecharacteristics by using discrete-event simulation methods.This presentation will briefly demonstrate what the timed automaton is, using asimple example, and will discuss some issues to consider when translating thepaper-based model to a functioning computer simulation.

* * *

0950 Jiayi Zhou, M.Sc. Computer Science

Title: Interactive Application to Visualize Chemical Structural Similarity

Abstract: The recognition of chemical similarities between molecules plays an importantrole in chemical science, especially in the state-of-the-art subjects of chemistry,biology and pharmaceutics. Traditional methods of structure recognition are timeconsuming, usually involving a lot of experimentation and computational effort.The goal of my research is to create an application which is able to compare twochemical structures automatically and provide an interactive way between theuser and the application in order to obtain the best result. My application willdisplay the structure in 3D. In addition, the application is able to display twogiven chemical structures at the same time. Furthermore, the application’s mainfeature is recognizing the chemical similarities of the given two structures andallowing the user to manipulate the structures in 3D.

* * *

1020 Pranjal Patra, M.Sc. Computer Science

Title: A Novel Method to Facilitate Biological Interpretation of High ThroughputExpression Profiling Data

Abstract: High-throughput expression profiling techniques have gained significantpopularity over past years. The goal of the project is to facilitate the biologicalinterpretation of results from expression profiling experiments. In particular, wewill provide researchers with gene sets (for example pathways) associated to thebiological phenomenon under study and their likely regulators. To achieve thisgoal we will build a system where users can input the gene expression profile datafrom their expression profiling experiments and get a list of disturbed gene setsand inferred transcription factors that possibly regulate these gene sets. In additionto the data from the expression profiling experiment, the system will also need adatabase of gene sets and quantitative information about the binding of specifictranscription factors in the regulatory regions of genes.

* * *

1040 Sadegh Ekrami, M.Sc. Computer Science

Title: Find All Possible Bindings in Colored Petri-nets

Abstract: Formal models are considered as an intermediate stage between the design andimplementation phase of a system. They are useful for simulating, testing,modeling and demonstrating system’s behavior in micro and macro scale. Petrinetsare one of the popular formalism which has been proposed by Carl A. Petri in1962. Many improvements have been proposed by modelers and significantchanges have shaped this powerful formalism in a way that presently manyscientists, practitioners and system designers are using for their work. ColoredPetri-nets (known as CP-nets) is a graphical formalism for design, specification,simulation and verification of systems. CP-nets fold an ordinary Petri-net intosimpler one, eliminating similar parts of the model. Petri-nets are in particularwell-suited for systems in which communication, synchronisation and resourcesharing are important. In this proposal, a new approach to finding all possiblebindings for CP-nets is being outlined.

* * *

1100 Ryan Zier-Vogel, Ph.D. Computer Science

Title: Predicting RNA Secondary Structure Using A Stochastic Conjunctive Grammar

Abstract: In this presentation I will talk about how I extend a class of grammars calledconjunctive grammars to a stochastic form called stochastic conjunctivegrammars. This extension allows the grammars to predict pseudoknotted RNAsecondary structure. Since observing secondary structure is hard and expensive todo with today’s technology, there is a need for computational solutions to thisproblem. A conjunctive grammar can handle pseudoknotted structure because ofthe way one sequence is generated by combining multiple parse trees.I create several grammars that are designed to predict pseudoknotted RNAsecondary structure. One grammar is designed to predict all types ofpseudoknots and the others are made to only predict a pseudoknot called H-type.These grammars are trained and tested and the results are collected. I am able toobtain a sensitivity of over 75% and a specificity of over 89% on H-typepseudoknots.

* * *

1140 M. Jalal Ahammad, M.Sc. Computational Science

Title: Numerical Simulation of Multi-scale Flows in a Geological Reservoir

Abstract: Simulation of flow in reservoir is a challenging topic; it depends on wide varietyof physical conditions. We know petroleum resources are found within rocks thathave a sufficient interconnected void space. The actual flow of fluid occurs on asmall scale in the void space between rock grains. On the other hand, the rockzones are extended to several kilometers in the lateral directions. A subsurfaceflow is a multiscale problem that is characterized by a wide range of length andtime scales. Typically, a characteristic length scale for the pore size in a porousformation is O(micron) and that for a reservoir is anywhere between O(100m) toO(km). Therefore, the numerical simulation of a multiscale flow in a salineaquifer or in an oil reservoir requires O(1027) grid points. Capturing such a flowfrom the pore scale to reservoir scale is thus a challenging endeavour. Therefore,very often it is difficult to do a proper qualitative and quantitative simulations.To resolve this problem, one can use some upscaling method, but most of theupscaling method have some limitations.We use volume-averaging techniques and lagrangian approach to address thisproblem

* * *

1200 Grigory Vorobyev, Ph.D. Computer Science

Title: Conformity and Nonconformity in Collective Robotics

Abstract: This work concentrates on the social phenomena of conformity andnonconformity in the context of collective robotics. As a case study, we refer tothe collective sorting task, where a group of relatively simple robots gatherobjects of different types into homogeneous clusters. Swarm robotics offersdistributed algorithms for solving this problem. While swarm-based algorithmstypically show slower convergence than centrally-controlled approaches, theiradvantages are simplicity, flexibility and robustness. Swarm agents are extremelysimple and unaware of each other. In this work, we extend the existingswarm-based approach by introducing additional social information to theoriginal algorithm. We show that increasing social awareness of the agents andawarding them with the ability to choose between conformity and nonconformityboosts the overall performance of the collective sorting task significantly, whilemaintaining the flexibility and robustness of the approach.

* * *

1220 Baharehsadat Pourazar, M.Sc. Computer Science

Title: Real-time Depth Estimation from Stereoscopic Video for Augmented RealitySystems

Abstract: Augmented Reality systems combine standard video input with computergeneratedobjects or environments. Stereoscopic video is a unique type of videosource, which is not typically used in Augmented Reality systems. In addition tocolor, stereoscopic systems allow viewers to perceive depth, thus providingimportant information to support navigation and manipulation tasks. Usingparallax, stereo vision systems determine depth from two or more images whichare taken at the same time from slightly different viewpoints. In this research wepropose to combine stereoscopic systems in an Augmented Reality system. Themost important and time consuming task for stereo vision systems is theidentification of corresponding pixels within a pair of stereo images. One of themost important challenges in this research is to calculate depth accurately and inreal-time using left and right input streams of video. After depth information hasbeen extracted from a scene, it can be transmitted over the network to remoteusers for tele-collaboration or it can be combined with 3D models to generate anAugmented Reality environment to be displayed on regular screens or on 3D seethroughglasses or displays.

* * *

1240 Scott Watson, Ph.D. Computer Science

Title: The Unfulfilled Potential of Video Game Agents

Abstract: Video game agents are often placed in dynamic, high-dimensionalenvironments with partial information. These conditions are fertile ground forthe application of advanced artificial intelligence techniques. Despite thegrowing importance of the industry, the state of the art of commercial videogame AI remains stubbornly primitive. In this talk, I will discuss the variousdifficulties research in this field poses, the limitations of the currentlydeployed solutions and outline several techniques and concepts that I believemay have the potential to deliver superior results including behaviourimitation, agent model sharing, action composition and cultural algorithms.

* * *

1400 Md. Monjur-ul-Hasan, M.Sc. Computer Science

Title: Event Based Anomaly Detection from Multiple Geospatial-Temporal Datasets

Abstract: The increasing rate of geospatial-temporal data calls for geovisual analyticsapproaches to extract important and hidden information from them. Theseapproaches help analysts to perform reasoning about the data without readingdetails. Most of the geovisual analytics systems focused on similarity analysis,whereas fewer systems focused on anomaly analysis. All these existing systems,data elements are considered as the unit of analysis. In many cases, data elementsof an individual dataset can be grouped based on common criteria such as sametime, location, or subject. Each such group is called an event. The problem ofdetecting events from geoapatial-temporal datasets, and detecting anomaliesamong these events are addressed as a part of geovisual analytics research. Thispart of the work intends to detect events from movement, and point sample data,utilizing those events to support the analysis of different types of anomalies. Theoutcome may help the development of geovisual analytics systems that considerevents as unit of analysis. The detection of different types of anomalies amongevents may also help the development of geovisual analytics systems from theperspective of anomaly analysis. As a case study, datasets regarding fishingactivities off the coast of Nova Scotia are used. This data was collected byFisheries and Ocean Canada during 2008 – 2009. The vessel movement logs, andlandings reports are considered. The vessel movement logs describe where thevessels went in different times of their fishing. This is an hourly measurement oflocations of fishing activities. The landings reports contain reported locations offishing activities. From these datasets, fishing events of different vessels aregenerated. Each of these events has two types of anomalies: missing data pointsand location mismatch. A fishing event has missing data point anomalies, if oneor more movement data points are missing in it. The detection of locationmismatch anomalies is complicated and requires user interaction. For a givenevent, this type of anomaly detection involves matching the fishing location in thelanding data and movement locations in the movement data. The distancedetermination between a reported location and a vessel movement path is itselfchallenging. In addition, a user configurable threshold is required, whichdetermine whether the calculated distance is considered as an anomaly in thatevent. All these challenges are addressed in this work as well as an event basedanomaly detection system is also provided.

* * *

1420 Md Mustafa Kamal Bhuiyan, M.Sc. Computer Science

Title: Image Search Results Organization Based on Metadata and Visual Features

Abstract: Effective image search results organization facilitates the searchers in theirsearching activities. Traditionally image search results are rganized in paged gridlayouts, where the results are ranked based on the relevance to the query.However, the paged grid layout does not consider the valuable high-level or lowlevelfeatures of the images such as metadata or visual features. It is also static innature, where the searchers have limited ability to explore and manipulate theimage search results. To overcome the problems of paged grid layouts, manyresearchers have explored organizing images based on visual features or textualinformation associated with the images. Since, the image search results usuallycontain different levels of semantic, visual, and temporal features, the effectiveorganization based on these features can facilitate the searcher's exploration tasks.However, it is challenging to extract feature vectors from the metadata by whichimage search results can be organized effectively. Metadata and visual featurebased image organization, Mavio: is a system, which is organizing the imagesearch results based on the combination of metadata features (i.e., user-generatedtags and device-generated timestamp) and visual features. In this system, a novelmethod is developed to extract the feature vectors from the metadata associatedwith the images. This system is collecting images, tags, and timestamps fromFlickr, and then produces feature vectors based on metadata and visual features.These feature vectors are used to organize the image search results. This systemalso includes a dynamic weight adjustment feature by which the searchers will beable to reorganize their search results on the fly based on the different importancelevels on the different features. This system will provide a highly interactive searchinterface, where the searchers will be able to organize and explore the imagesearch results.

* * *

1440 Yoones Asgharzadeh, Ph.D. Computer Science

Title: SESM: Semantic Enrichment of Schema Mapping

Abstract: Schema mapping is becoming pervasive in information integration throughdata exchange and data integration. In this paper, we show that current schemamapping generation and verification techniques are not capable of reconcilingsome semantic heterogeneity because of ambiguities in interpreting differenttypes of relations. To address this problem, we enrich the mapping generationprocess using conceptual models to recover the semantics of relations. Thetechnique we propose in this paper not only avoids generating ambiguousmappings, but also generates some new semantic mappings that are neglected inmany schema mapping techniques.

* * *

1500 Akhilesh Mishra, M.Sc. Computer Science

Title: First-Person Perspective Gesture Recognition

Abstract: Gesture recognition has gained a lot of popularity in the fields of ComputerGraphics and Human Computer Interaction. Gesture interaction is an easy andeffective way of interacting with a computing system. Nowadays devices likeMicrosoft's Kinect, Nintendo's Wii and Sony's Move make use of gestures forgame play. However, gesture interaction is not limited to gaming applications andit can be used for general purpose interactions as well. The idea behind thisresearch is to develop a first-person perspective gesture recognition module for aTelecollaboration and telecommunication system. Existing gesture recognitionmethods are primarily based on a third-person perspective, i.e., the camera or thegesture-recording device is in front of the user, and the computational analysis ofthe player's gestures is done from the perspective of an external viewer, or thirdperson.The main disadvantage of third-person gesture recognition is that the userwould always have to be present in front the system and thus the system is notmobile. First-person gesture recognition would allow the user to be mobile andcould be used in conjunction with Augmented Reality systems to improve theoverall user experience. To test this hypothesis, the viability and efficacy of thefirst-person gesture recognition module will be evaluated through a user study.

* * *

1530 Md. Asaduzzaman, M.Sc. Computer Science

Title: A Model Cloud Computing Approach Ensuring Transactional Consistency

Abstract: Mobile devices are getting smarter day by day in terms of capabilities andcomputing power. This triggers rapid growth of mobile applications which hasgained momentum using cloud computing infrastructure. Cloud computing offersminimum management efforts for mobile applications at a very low cost. In thistalk we will focus on a mobile cloud computing approach in Intelligent TransportSystem (ITS) which will ensure minimum waiting time for emergency vehicles intraffic intersections. Here we will discuss the research challenges (e.g.,consistency, security, and privacy) and propose a transactionally consistentalgorithm integrating transactional and cryptographic primitives.

* * *

1550 Anastasia Gurinovich, Ph.D. Computer Science

Title: Dyadicity and Heterophilicity in Complex Networks

Abstract: In this presentation I will discuss two interesting parameters, i.e. dyadicity andheterophilicity, which measure how the nodes in a graph/network are connectedto each other with regard of having a certain common property. In many truenetworks (e.g. protein-protein interaction networks or mobile phone networks),nodes sharing the same property tend to be connected to each other more thanexpected for a random configuration. Thus, dyadicity and heterophilicitymeasurements could potentially be applied to evaluate inferred co-expressionnetworks on whether they have true underlying biological structures.

* * *

1610 Momeneh Taban, M.Sc. Computer Science

Title: A Recommender System for Breast Cancer Patients

Abstract: An ongoing challenge in the information age is finding information relevantto a particular need. One area in which this is particularly problematic is themedical domain, where patients suffering from certain conditions seek advice onmanaging their health. Personalized recommendations can be useful in thiscontext. A recommender system can assist users to locate relevant informationand choose the best option that matches their needs.This thesis will concentrate on designing and developing a recommendersystem for patients who suffer from breast cancer. The system will be designed toprovide users with some behavioral, dietary and medication guidelines, as well asto provide its users with some information about products and services related totheir disease. The recommender system will use patients individual information(i.e. current health status, life style, personal information) to generatepersonalized recommendations.

* * *

1630 Ben Fowler, Ph.D. Computer Science

Title: Evolvability and Robustness in Genetic Programming

Abstract: Evolvability in genetic programming refers to the ability of an individual orpopulation of programs to produce higher fitness individuals. Individual fitness istypically judged by the ability of the individual program to solve a specificproblem by using all available test cases; such an approach encourages theevolution of programs which best solve a specific problem, though it does notencourage evolvability. Robustness in genetic programming refers to the abilityof an individual or population of programs to retain functionality despiteperturbations which occur during evolution. Robustness competes withevolvability, but both qualities can facilitate faster evolution. By effectivelymeasuring both robustness and evolvability we can generate individuals with bothproperties, or at least analyze the tradeoff between them. This will lowerthe computational cost of genetic programming, which has many practicalramifications. I propose to develop new metrics for evolvability androbustness. Previous metrics required analysis the fitness of all possibleoffspring of each individual, but this is computationally expensive, and isexponentially so, if we intend to consider multiple generations. Novelmetrics considering the structure of an individual, rather than the fitness of itsoffspring, will be developed for this research. These metrics will be exploitedduring fitness evaluation to allow more evolvable and robust individuals to beselected with a reasonable increase in computational cost, yielding a decrease incomputational cost overall due to the reduction in required number of generations.

* * *

1710 Esteban Ricalde-Gonzalez, Ph.D. Computer Science

Title: Panoramic Review of the Wind Speed Forecasting Problem

Abstract: In the last years, wind power rapidly grew as an important renewable energysource. However, efficient wind speed forecasting methods are required in orderto properly fulfil the operating, planning and supplying requirements of theelectric industry.This talk gives a general description of the wind speed forecasting problemand presents a review of different solution routes explored in the literature,with emphasis in the machine learning approaches of time-series models.

* * *

1730 Liuhua Zhang, M.Sc. Computer Science

Title: Improving Specificity in Mammography Using Cross-correlation betweenWavelet and Fourier Transform

Abstract: Breast cancer is in the most common malignant tumor in women. It accounted for30% of new malignant tumor cases. Although the incidence of breast cancerremains high around the world, the mortality rate has been continuously reduced.This is mainly due to recent developments in molecular biology technology andimproved level of comprehensive diagnosis and standard treatment. Earlydetection by mammography is an integral part of that.The most common breast abnormalities that may indicate breast cancer aremasses and calcifications. Previous detection approaches usually obtain relativelyhigh sensitivity but unsatisfactory specificity. This investigation was undertakento improve specificity.We investigated an approach that applied the discrete wavelet transform andFourier transform to parse the images and extract global statistical features suchas the mean intensity and the skewness of the intensity. Features were rankedbased on their entropy and the best 14 were used in the classification process.Four classifiers, including Back-propagation (BP) Network, Learning VectorQuantization (LVQ), Linear Discriminant Analysis (LDA), and Naïve Bayes(NB) Classifier, were tested and have achieved promising results. We arecurrently determining which combination of classifiers will provide optimallyhigh specificity.Keywords: mammography, wavelet transform, Fourier transform, Bayesianclassifier, specificity

* * *

1750 Norbert Obiekwe, M.Sc. Computer Science

Title: Digital Message Authentication Using Digital Signature Recognition

Abstract: Digital message authentication and verification is an actively growing area ofresearch. Currently there are various personal authentication techniques such asvoice, finger prints, and face and retina recognitions [1, 2, and 3]. These methodsof authentication are based on different personal psychological and behavioraltraits generally known as biometrics [4]. Moreover, the driving force behind thisactive authentication trend is above all, the ever growing role of internet anddigital transfers in today’s modern society. By authenticating digital messageswith a digital signature, the receiver can be sure that the message is from theexpected sender and hasn’t been modified by a third party (ensuring messageintegrity). Also, using a digital signature authentication method has significantadvantage over traditional authentication (such as passwords, PIN number, smartcards) due to the fact that digital signatures are not easily transferable, are uniqueto every person and can’t be stolen or lost. Above all, the process of signing andverifying a digital signature is by using complex encryption and decryptionprocesses and hashing key functions.The overview of signing, verifying and applications of digital signature are to bepresented to the forum.

* * *

1820 Syed Nasir Danial, M.Sc. Computational Science

Title: Simulated Human Behavior

Abstract: The technological advancement is a key factor of today life. Now we use complexintegrated machines which consist of millions of elements Such complex systems,although designed with great care, are always susceptible to errors particularlywhen relied on human to operate. Therefore, human-machine interface remain acritical link to ensure safety of the system and also safety of the operator. Onepossible mechanism to improve safety of such interface is to train human forexpected and unexpected scenarios, if possible in virtual scenarios. This studyattempts at modeling an artificially intelligent agent which is supposed to embodythe knowledge of a specific working environment, such as an oil rig, where theagent faces harmless as well as potentially hazardous situations. The target is todecide which actions or plans the agent should come forth in order to mitigate anycurrently observed situation. The whole work is divided into three main layers,viz., (a) the information gathering layer (layer-a), (b) the knowledge baseddecision layer (layer-b), and, (c) the action selection layer (layer-c). The focus ofthe present study is on layer-b, where a knowledge base which exploits theconcept of frequency gambling and similarity matching is proposed. The layer-ais responsible of observing the agent’s environment for any potential hazardoussituation, such as a fire event, and then observing the surroundings to be used inlater decision related stages in the layer-b. The layer-a is also responsible fordifferentiating among various possible hazards, such as fire, smoke, andexplosion. The decisions taken in the layer-b are translated into set of actions inthe layer-c, where a set of actions is formed by selecting appropriate actions fromthe list of all possible actions the agent can perform, such as arriving at a target,chasing a target, following a specific path, etc. This way, the agent is supposed toachieve a task by first analyzing the problem (layer-a), then planning the courseof actions (layer-b) and then acting upon the decisions made (layer-c). Finally, theagent also monitors the aftereffects of the actions under any current setting andupdate its knowledgebase on the bases of estimating if the actions were reallymeaningful in mitigating the situation or not. The work has a profound scope inmany pragmatic settings, such as development of intelligent trainers and otherexpert systems.

* * *

1840 Tamkin Khan, M.Sc. Computer Science

Title: AUV Mission Planning using SAT/SMT Solvers

Abstract: Autonomous underwater vehicles such as SLOCUM gliders have been usedfor many tasks, in particular underwater surveys, for a number of years. During amission, several factors impact their performance: ocean currents, lack ofcommunication and GPS while under water, weather, etc. At present timeengineers do not seem to do much mission planning: they just give a fixed list ofway-points to an AUV before sending it on a mission. This approach is becominginefficient as present-day AUVs can travel over a long distance. SLOCUMgliders are small and not too powerful, but they can stay in water for over amonth, and a group of gliders working together can visit a significant number ofway-points. Developing techniques for planning such missions for possiblyheterogeneous groups of gliders is the goal of this project.We can formulate this problem as an AI planning problem, and model it as aweighted directed graph, where nodes correspond to states for differentgliders and edge weights are transition costs from one state to another. Weassuming that this graph will be generated from ocean current (NetCDF)historical data file or a numerical model of ocean currents. But searching for anoptimal routes and schedules for each glider is an NP-Hard problem. There are noknown algorithms guaranteed to solve it efficiently; however, there are heuristicsolvers (SAT/SMT Solvers) that can perform well in practice. In this project, wewill investigate whether these solvers can be adapted to the AUV scheduling androuting problem.Much of the work intended to be done is encoding our problem as an input tosuch solvers, and analyzing their performance. Since the performance usuallydepends heavily on the encoding, we might be trying a range of encodingmethods.

* * *

1900 Cyril Michael, M.Sc. Computer Science

Title: Evaluation of Wireless Mesh Networking with Xbee

Abstract: In this project we are trying to do a performance evaluation of ZigBee meshnetworks based on XBee modules, in terms of packet delay and received signalstrength (RSSI). We evaluate the performance for both line-of-sight (LOS) andnon-line of sight (NLOS) environments in buildings and a parking lot. We are alsoevaluating the Packet Delay in single-hop and multiple hops by sending smallpackets of data every 5 seconds, and wait for the acknowledgment of that packetto the source. Packet Delay is actually measured and compared in single-hop andmulti-hop networks. The results of our study are useful for building wirelesssensor networks and communication networks of mobile robots where thecommunication distances between nodes are greater than the single-hop range andhence communication between nodes requires multi-hop transmissions.

* * *

1920 Wagdi Alrawagfeh, Ph.D. Computer Science

Title: Norm Representation, Normative Reasoning and Agents Behavior

Abstract: Norms play an obvious role in coordinating, regulating and predicting agentsbehavior in open multi-agent societies. Much work has been done on modeling anddeveloping normative multi-agent systems. Nevertheless, little work has dealt withthe impacts of norms on an individual agents practical reasoning. Norms in openmulti-agent societies are not fixed, they might emerge, change or vanish; thereforeagents need a mechanism to adapt their behavior accordingly. We propose a normrepresentation and a normative reasoning strategy to help BDI agents to reason atrun time about the best behavior they may choose. Our technique has been appliedto a simple game simulation called mineral mining. Jason is a BDI interpreterwhich has been chosen to implement our work.

* * *

1940 Closing Remarks

Computer Science
|
Faculty of Science

Computer Science
|
Faculty of Science

Student Research Forum 2013

Tenth Annual
Student Research Forum

Student Research Forum 2013

Tenth AnnualStudent Research Forum

Tenth Annual
Student Research Forum