NTU From left: Asst/P Lin Shang-Wei, A/P Adams Kong,
Asst/P Liu Yang, A/P Ng Wee Keong, Asst/P Alwen Tiu
and Asst/P Xing Zhenchang.

Digital Crime & Forensics

Researchers in the group study computational methods for digital crimes and forensics. They are developing new biometric traits e.g., skin marks, androgenic hair and blood vessels hidden in color images, and algorithms to identify criminals and victims in evidence images, where their faces are not observable. The target evidence images can be taken from child sexual offenses e.g., child pornographic images, other sexual offenses, terrorist activities and riots. Some techniques developed by the group were applied to a child sexual offense case in U.S. The group worked with U.S. Department of Justice and U.S. Immigration and Customs Enforcement. They are working closely with Singapore Police Force. In addition to the development of the new biometric traits, they also study security risks in identity systems for preventing digital crimes.

Research Areas

  • Biometrics
  • Forensics
  • Pattern Recognition
  • Image Processing

Cyber Security

The Cyber Security group’s research interest focuses on the development and application of techniques to enhance high assurance systems’ security, emphasising on the use of formal methods to provide system foundations correctness. The Cyber Security group has active projects in collaboration with Singapore’s governmental agencies and companies to provide effective solutions to create a secure cyber space.

Research Areas

  • Malware Modelling, Detection, and Classification
  • Mobile Security
  • Vulnerability Modelling and Detection
  • Security Protocols Analysis
  • Binary Code Analysis
  • Formal Methods on Hardware Modelling and verification
  • Hypervisor/ Micro-kernel Modelling and Formal Verification
  • Performance Analysis
  • Reliability and Trustworthiness Analysis

Computer Networks & Communications

NTU First row from left: A/P Lee Bu Sung, Francis, A/P Lau Chiew Tong and
A/P Dusit Niyato. Second row from left: A/P A S Madhukumar,
A/P Yeo Chai Kiat, A/P Wang Ping and A/P Luo Jun.
Not in photo: A/P Li Mo and A/P Wen Yonggang

The Computer Networks & Communications research group has research focuses in the areas of sensor networks, IoT, data center networking, software-defined networks, and 5G networks. The research groups aim to address the system design and resource optimization issues to achieve certain objectives under different constraints. In sensor networks and IoT, the energy and communications resources are scarce and their usage has to be jointly optimized. In data center networking and software-defined networks, new network optimization schemes such as bandwidth allocations to support various applications, e.g., multimedia streaming have to be devised. In 5G networks, new system design to support advance transmission technologies, e.g., massive MIMO, has to be developed to meet the expectation of the new mobile applications. These problems have never been studied before, and new development can achieve significant contributions.

Research Areas

  • Sensor networks
  • IoT
  • Data center networking
  • Software-defined networks
  • 5G networks

Computational Intelligence

Researchers in Computational Intelligence (CI) research group investigate natural and artificial systems to comprehend principles that render intelligent behaviour possible in complex changing environments.

Contents

The CI research group core research is focused on devising intelligent machines capable of learning, understanding and reasoning about past actions, so as to provide creative solutions to real world complex problems.

Research Areas

  • Cognitive and Neuro Systems
  • Decision Support Systems
  • Evolutionary, Memetic and Statistical Learning
  • Fuzzy Systems
  • Intelligent Multi-Agents
  • Intelligent Systems and Devices
  • Machine Learning
  • Nature-Inspired Systems
  • Natural language processing
  • Sentic computing

NTU First row from left: Prof Ong Yew Soon, A/P Quek Hiok Chai,
A/P Tan Ah Hwee and A/P Miao Chunyan.
Second row from left: Asst/P Kelly Ke Yiping, Asst/P Sinno Pan,
Asst/P Bo An, A/P Suresh Sundaram and A/P Zhang Jie.
Not in photo: Asst/P Erik Cambria

Biomedical Informatics

The mission of the Biomedical Informatics research group is to develop the capabilities of information technology to transform in many aspects of medicine and health care. It is also aimed to discover, encourage, train, and support the new-generation of medical scientists via graduate programmes to support biomedical engineering and related fields.

Research Areas

  • Bioinformatics
  • Quantitative Methods in Medicine
  • Diagnostic Instrumentation
  • Medical Imaging and Visualisation
  • Healthcare Information Systems
  • Telemedicine
  • Clinical Data Mining
  • Physiological Models and Computer Simulation

NTU From left: A/P Kwoh Chee Keong, A/P Lin Feng, Prof Rajapakse Jagath
and Asst/P Zheng Jie.

Hardware & Embedded Systems

The Hardware and Embedded Systems (HES) group undertakes research to overcome emerging challenges in embedded and high performance computing systems. The major goals of HES are hardware acceleration, reconfigurable computing, heterogeneous architectures, architectural translation of complex algorithms, system integration and design methodologies.

Research Areas

  • Secure and reliable embedded systems
  • Heterogeneous architectures and compilation tools
  • Cyber-physical systems
  • Embedded vision and autonomous systems
  • Overlay architectures for FPGA-based computing
  • Brain-computer interface
  • Synthesis for emerging technology
 

NTU From left: A/P Vinod Prasad, A/P Nicholas Vun, A/P Douglas Maskell,
Asst/P Anupam Chattopadhyay, Asst/P Lam Siew Kei and
Prof Thambipillai Srikanthan. Not in photo:
Asst/P Arvind Easwaran and Asst/P Nachiket Kapre

Data Management & Analytics

The Data Management & Analytics research group explores fundamental problems in managing and mining massive volumes of data that are of significance in the 21st century. It currently consists of 7 faculty members from the School of Computer Engineering in NTU. Members of the group have consistently published their novel research in top-tier data management and analytics venues and have attained excellent international

NTU From left: A/P Hui Siu Cheung, A/P Sun Aixin, A/P Cong Gao,
Asst/P Ho Shen Shyang, A/P Sourav Bhowmick and A/P Xiao Xiaokui.

visibility in these areas. They also contribute significantly to the international research community by serving as chairs, keynote speakers, and PC members/reviewers in leading international conferences and journals.



Contents

Research Areas

  • Data Analytics
  • Data Mining
  • Data Management
  • Information Retrieval
  • Information Security and Privacy

Multimedia

NTU From left: A/P Cai Jianfei, A/P Lin Weisi, A/P Chng Eng Siong, A/P Chan Syin (OPE), A/P Cham Tat Jen, A/P Deepu Rajan and A/P Chia Liang Tien.

The Multimedia research group is dedicated towards discovering breakthroughs in automatic processing and analysis of images, audio and video using intelligent computational systems, so as to be able to distill important high-level semantic information from such data. Besides traditional multimedia data, other types of data investigated include multi-view imagery, RGB-depth data and also microphone array audio.

Research Areas

  • Recognizing people, objects and scenes based on images
  • Accurately reconstructing the shapes of objects from multiple images
  • Localizing and analyzing the motion of objects across video sequences
  • Tracking people and recognizing their actions and activities
  • Segmenting and matting objects of interest to separate them from the background
  • Detecting salient parts of images that pertain to human attention
  • Conducting speech recognition robustly in noisy, uncontrolled conditions

Visual & Interactive Computing

NTU From left: A/P He Ying, A/P Alexei Sourin, Prof Seah Hock Soon,
A/P Goh Wooi Boon, A/P Zheng Jianmin and A/P Qian Kemao.

The Visual & Interactive Computing (VIC) group focuses on research and education in visual computing that addresses the interactions between humans, computers, and real & virtual worlds. Researchers in VIC are devoted to investigating, inventing and integrating the computational capabilities and their interactions with visual data from fundamental theory to practical applications and from physical systems to software development.

Research Areas

  • Computer graphics
  • Visualization
  • Animation
  • Human computer interaction
  • Games
  • Augmented & virtual reality
  • Haptics

Parallel & Distributed Computing

NTU First row from left: A/P He Bingsheng, A/P Anwitaman Datta,
Prof Cai Wentong and Asst/P Elvis Liu Sze Yeung.
Second row from left: Asst/P Fan Rui, A/P Hsu Wen Jing,
Prof Sun Chengzheng (OPE) and A/P Tang Xueyan.

The mission of the Parallel and Distributed Computing Group is to form a group of excellence in the areas of parallel and distributed computing. We research the fundamental problems and challenges in parallel and distributed computing, and emerging applications such as big data. The group has organised several leading international conferences in these areas. Particularly, it has gained an international reputation for its research on distributed simulation and collaborative technology. The group also strives to seek new industrial projects where parallel and distributed processing can provide a solution to real problems and foster research collaborations both nationally and internationally. Expertise of the faculty members in the group is internationally recognized and the group has also become an internationally recognizable brand-name for its research in the areas of large-scale simulation, collaborative computing, and high performance computing. Particularly, the simulation group is ranked 9th in the world and 1st in Asia according to the publication records for the past 10 years[1]).

Research Areas

  • Collective Technologies – Have many applications (e.g., collective design) and potential to increase productivity
  • Agent-based Modelling and Simulation Technologies – Have many applications (e.g., smart cities and urban analytics)
  • Systems and Tools on Emerging Parallel Architectures – Improve the efficiency and programmability of emerging hardware and support the execution of large-scale applications
  • Distributed computing – Improve the efficiency and reliability of distributed computing systems and support the execution of large-scale applications



Contents

Motivation & Problem

The physical nature of communal games involving face-to-face groups contributes significantly to the quality of game play and social engagement. However, designing interesting communal games on mobile phones is currently limited because phones are unaware of their spatial relationship with other phones. The premise of this research is that the availability of instantaneous spatial position and orientation information between co-located phones can create a very interesting design space for developing novel and engaging spatial-aware smartphone-based communal games.

The Approach

A specially-designed patterned surface called the gaming arena is used to provide visual cues to the smartphone’s camera. This patterned surface consists of regularly spaced colored visual markers that provide two important visual features that allow the smartphone to be spatially-aware, see Figure 1(left). Firstly, each square marker is uniquely and methodically colored, thus enabling a real-time vision algorithm within the Android-based phone to compute marker identity by processing the color video captured from the phone’s rear camera at 25 frames per second. Identity of the markers currently within the visual view of the camera allows the computation of the instantaneous position of the phone within the gaming arena. Secondly, the visual markers provide corner-like point features that the vision algorithms can extract in real-time. The PnP algorithm is then applied to these feature points to estimate the 3-dimensional (3D) pose of the camera with respect to the flat gaming arena. A client-server wireless WiFi network is setup between multiple phones to allow spatially-aware collaborative smartphone games to be played within this arena. A library of Android-based software APIs have been developed to provide 3D positional information when the built-in camera of the Android smartphone captures the instantaneous view of the gaming arena. These information include the 2D spatial position of the phone in the arena, its 3D pose (i.e. tilt), orientation and relative height from the floor.

NTUFigure 1. (Left) A gaming arena comprising of a PVC canvas printed with regularly patterned visual markers that is visible to the rear-facing camera of a smartphone. On the smartphone is a collaborative game called Blind Spots that allows 3 to 4 players to create formation based on the target displayed on their phone screen. (Right) Four 9-year olds playing Blind Spots during a field trial held in the research lab.

Applications

Based on this library of Android APIs, two collaborative games have been developed to date. The first is called Blind Spots. The design objective of Blind Spots is to create an engaging game play environment that encourages physical movement and the development of positive social skills such as communication, sharing, teamwork, group processing and leadership. The goal of the game play is for 3 to 4 players to organize themselves into appropriate formation within the gaming arena based of the location cues given on the screen display of selective smartphones during each round of the game, see Figure 1(right).

Scoring is based on how close each team member is to the target location when a timer expires. This game was trialed with three different groups of young participants, with one group playing the game twice over a 5 month period. All participants gave a resounding “Yes!” when asked if they enjoyed playing the game, even children who played the game twice. Two features of the game stood out for most players; they were the teamwork element of the game play and the physicality of running around.

NTUFigure 2 –The positions of the players on the gaming arena determine the selection of their number in a corresponding virtual numeric grid layout (denoted by different colored squares on the phone display). In the example display on the left, the four selected numbers 1, 1, 4 and 5 add up correctly to the target value 11, which indicated as a circle on the top-right corner. The other game called Number Bonds was designed to create an engaging play environment to teach arithmetic addition (e.g. number bonds) through physical movement and collaborative group processing. The goal of the game is for 3 to 4 players to work together in selecting the right combination of numbers on

NTU

their phone screen that will add up to the value of the target number given to them during each round. The selection of the number square within the 3×5 numeric grid layout on their phone display is based on the corresponding physical position of the player in the gaming arena. As such, the group has to work together to decide what combination of numbers they wish to make and quickly move to the respective positions in the gaming arena before a timer expires. Again, all participants who played Number Bonds during the trials said they enjoyed the game. The game features the children found most enjoyable were the need to communicate, cooperate and play as a team. We see potential for primary schools to adopt the Number Bond game as a physically fun way of rehearsing tiresome number bond exercises, which children currently do using repeated worksheet practice.

Point of Contact

GOH Wooi Boon
aswbgoh@ntu.edu.sg
This project is funded by a MOE Tier 1 grant



Contents

Introduction
Terrorist Attacks in large public events can cause catastrophic effects, e.g., the Boston Marathon bombings in April 2013. However, it is challenging to develop efficient strategies to secure such public events due to their dynamic nature, i.e., an attack can occur at any time during an event, while the impact of attacks at different locations changes over time. Given the fixed and limited number of security resources, dynamic security strategies in which a security resource can be relocated among potential targets during the event should be considered.

NTU

In order to address the challenges of computing optimal security resource allocation for protecting large scale public events, we combine techniques from computational game theory, optimization, and artificial intelligence. We aim to develop novel models and efficient algorithms for the optimal randomized deployment of limited security resources to deter and prevent terrorist attacks to large public events that balance available resources, terrorist surveillance/reasoning capability, and dynamics during large public events. Specifically, we investigate fundamental research issues including building the game-theoretic model for protecting public events, designing efficient algorithms for solving the security games, and enhancing solution robustness against uncertainty and bounded rationality of adversaries.

Motivation Domain

On April 15, 2013, two pressure cooker bombs exploded at 2:49 PM EDT during the annual Boston marathon. The bombs exploded about 12 seconds and 210 yards (190 m) apart, near the finish line on Boylston Street, Boston. They killed 3 civilians and injured an estimated 264 others. In addition to the direct loss such as death and injuries, such successful terrorist attacks can induce huge fear in the general public and can generate huge long term cost to the economies of the countries and target cities.NTUIntelligent deployment of limited security resources (e.g., police) to protect such public events is therefore extremely important and difficult since the importance of targets changes over time. For instance, the value of targets (areas) along a marathon track changes over time, since participants and spectators move along the track during the event. We can divide the whole marathon course into many small segments to represent many potential attacking targets. As the public event proceeds, the number of participants and spectators near the starting line will decrease, thus the damage caused by attacking targets (areas) around the starting line will thus decrease. Similar variations in the value of targets happen for targets along the way to the finishing line.


A target near the finishing line is critical when the marathon race is approaching the end. Intuitively, if the total number of security resources (e.g., police) is fixed, it may be better if the defender transfers some security resources from targets near the starting line to targets near the finishing line as the event proceeds.

Thus a dynamic allocation of security resources during the event is in need and is more efficient. Furthermore, the low frequencies of public events like annual marathons make it difficult for the attacker to conduct surveillance and respond. Thus, a pure dynamic resource allocation strategy that minimizes the worst case loss of the defender is a robust solution.

Progress

While there have been many existing works on applying computational game theory and artificial intelligence for security, none of the existing works considered dynamic security resource allocation for protecting large public events in which the values of different targets will dynamically change. Recent work has successfully deployed game theoretical models/algorithms to scheduling resources for many real-world security applications, including the ARMOR system for the Los Angeles International Airport Police, the IRIS system for the United States Federal Air Marshal Service, and the PROTECT system for the US Coast Guard. At the heart of these deployed systems are efficient algorithms for solving the games and dealing with uncertainty in the real world. Unfortunately, none of the existing models and algorithms can be applied to model and solve the public event security problem.

We thus have designed a new security game model with continuous strategies for both the defender and attacker to describe dynamic patrolling in domains with dynamic payoffs. We proposed algorithms to compute the optimal dynamic allocation of security resources under the zero-transfer-time assumption, i.e., a security resource can be transferred from one target to another instantaneously. The algorithms use continuous functions (e.g., piecewise linear) to represent targets’ utilities, so that they can solve the game without traversing the whole defender strategy space. We have also proposed an algorithm to deal with general cases where transfer of security resources takes time. The algorithm first constructs a game with discrete defender strategy space which is equivalent to the original continuous-strategy-space game, then solves the constructed game. Experimental results show that our algorithms significantly outperform existing approaches in the literature. Some results have appeared at AAAI’14.

Point of Contact

Bo AN
boan@ntu.edu.sg




Contents

Motivation and Problem

Common haptic interaction assumes that frequent collision detection of the so-called Haptic Interface Point (HIP) with 3D object models has to be performed. It requires the user to model the 3D scene with a high degree of visual realism, which is time consuming and not always even possible task. Moreover, number of polygons for haptic interaction is limited since it requires 1000 Hz refresh rate.

We propose image-inspired haptic interaction with 2D photos which are augmented with invisible haptic models. We believe that for many practical applications such haptic interaction with the augmented image will deliver a believable tangible immersion into the image. Our hypothesis is based on the fact that replacement of 3D visualization with image-based visualization is actively used in interactive panoramas, street walkthroughs, and online shopping. We add new modality to image-based visualization and achieve further immersion into the images by blending visual and tangible experiences within the displayed scene. The image will serve a purpose of guiding the user towards the parts of the 3D scene where haptic interaction has to be done while augmenting and correcting the overall visual-haptic perception of the scene. Therefore, “What You See is What You Touch” is used as an interaction principle.

The Approach

Haptic interaction with two-dimensional images is commonly performed either by calculation of haptic forces directly from the image or by reconstruction of whole 3D geometry from the image and using it for haptic interaction. We contributed in the past to the first group of methods where we proposed how to use pixelwise image intensity for calculation of the haptic feedback forces. In this project, we consider an approach alternative to the computationally demanding and ambiguous 3D geometry reconstruction. To haptically interact with the images, as if they were actual tangible 3D scenes, we augment the images with invisible haptic models matching some parts of the images essential for haptic interaction. These models may range from geometric surfaces and solid objects defining both visible and inner haptic properties of the objects, to 2D or 3D force fields which may have no any concrete visual appearance in the image (e.g., air or fluid flow, gravity force). Such haptic models we define in an additional “haptic” file, used together with the original image in a way how movie subtitle files are added to video files for displaying closed captions. Using our tangible images player, the multisensory interactive experiences can be sensed as if the actual 3D scenes we behind the images.

We do not reconstruct 3D model of the scene displayed in the image with proper depth location of the objects. Instead, we place the haptic models in front of the image plane while still preserving mutual occlusions wherever they happen. The user will have a feeling of an ability to reach with the virtual hand any part of the scene made tangible without performing a virtual walk-through the scene to reach them.

Besides this, we do not derive the parameters of the central projection from the image which is commonly done for accurately reconstruction of 3D objects. In fact, it may not be even possible with absence in the image of some reference objects with known sizes and orientations.

Instead, we interactively create haptic models which geometry is distorted by the central (perspective) projection so that their contours will match the object contours in the image when projected with parallel orthographic projection. While these models are placed in front of the image plane, the position of the HIP is projected onto with parallel orthographic projection. With reference to Figure 1, to match the contours of the distorted cylinder and cube, we use a truncated cone with non-parallel end-caps and a 3D object bounded by 6 non-parallel polygons. As a result, since haptic perception with one HIP is performed by mental interpolation of haptic feedback following the HIP moving along only one contour/ surface at the time, this approach provides us a believable combination of visual and haptic immersion into the image implementing the principle “What You See is What You Touch”.

We mostly rely on so-called FRep functions f(x,y,z)≥0 for defining haptic models since they allow for defining solid models, generalizing collision detection for different types of haptic models, supporting Boolean operations, and, finally, providing for very small size of the models since only defining function coefficients and parameters have to be stored in the haptic file.

NTU Figure 1. Location of haptic objects with reference to the image plane.

We define haptic models by drawing just a few strokes. In contrast to common sketch-based modelling, we do not aim to exactly match the contours of the visible objects since the mnodels will not be visually displayed. Our goal is to deliver a believable haptic perception of the tangible parts of the image. It requires different precision and rendering methods depending on the type of the haptic device used. For example, to have a solid feeling of the surface or contour where it is displayed in the image, the haptic model of the surface may be slightly displaced or enlarged due to elasticity of the haptic feedback force which will be delivered by the device. The haptic models are defined as Boolean combinations of some simple parts such as planes, cones, ellipsoids, etc. which are invariant to central projection distortions.

Haptic forces calculated from colliding with the haptic model are composed by normal components, which are functions of the depth of penetration of the HIP inside the model and tangential components, which are functions of the surface friction assigned to the haptic model. Once the force vector is thus computed, a texture component is added to the surface of the model to perturb the force vector. We did it by combination with the force vector obtained directly from the image.

Figure 2 illustrates haptic interaction with photos. 3D haptic force vector (red arrow) is projected onto the images to display how the interaction with images goes. The vector toggles off for the invisible surfaces of the models.




Contents

Significance and Impact

We expect that attention of researchers will switch to multisensory image interaction, especially tangible communication. There were multiple signs recently that several large vendors are considering tangible interfaces on mobile devices and personal computers (e.g., Apple patented “Touch-based User Interface with Haptic Feedback”, Samsung called for Global Research Outreach program on “3D Haptic Interface with Hologram Display”, Microsoft is researching on haptic keyboards). With our preliminary work done, we are ahead of other researchers who may embark onto similar projects, and we are ready to expand our ideas to these future devices.

Commercialization Potential

The proposed haptic interaction with images can be used in many practical applications ranging from virtual museums and cultural heritage applications, where images of the exhibits can be haptically explored without the need of a detailed 3D reconstruction, to common human-computer interaction, training and simulation where images and video of real or virtual scenes can be used as a source of haptic interaction with the objects displayed there.

This is especially useful for medical training applications. Many surgical operations are performed nowadays using minimally invasive techniques where the surgeon sees the surgical field as an image on a video monitor while operating remotely controlled surgical instruments. Given that the cost of 3D reconstruction of each surgical case can be very high, using examination videos for creating tangible panoramic images can lead to much faster and efficient simulations.

NTU

NTU Figure 2. Haptic interaction with images augmented with haptic objects. Direction of the force feedback vector is displayed as a red arrow.

Contributed Authors

Alexei SOURIN, assourin@ntu.edu.sg
ZHANG Xingzi
Shahzad RASOOL

iMASON: Towards Influence-driven Multi-level Analysis
of Online Social Networks

Social network is a social structure of nodes that are tied by various kinds of relationships, such as kinships, friends, web links, colleagues, citation links, etc. Since last decade, large on-line social networks have become very popular among web users. A key feature of many online social networks is that they are driven by social influence between users. Specifically, the structural and semantic properties of individuals, communities and network can be traced back to the social influence effect. Hence, studying the social influence between users in a network is of great importance in many applications such as viral marketing, online advertising, advanced web architecture design, etc.

In this research, we explore a framework called iMASON (influence driven Multi-level Analysis of online SOcial Networks) that systematically investigates social influence effect in social networks from multiple granularity including individual-level, community-level and network-level. Specifically, arrays of novel influence-related problems are tackled and many new interesting findings are uncovered in this research. For instance,

at the individual level we propose a novel algorithm called CASINO to evaluate each node’s influence and conformity. At community level, we propose AffRank to rank the ability of product communities to absorb new members which is driven by social influence effect. At network level, we propose a novel algorithm called CINEMA to solve conformity-aware influence maximization problem. We are currently extending our study to investigate these problems in the presence of competition as well as in dynamic networks.

Principal Investigator

Sourav S BHOWMICK

Collaborator

HUI Li (Xidian Univ, China)




Contents

Mobile devices, such as smartphones and tablets, are gaining popularity and increasingly being used as the main devices for communication, to store sensitive personal information, to access the web and to conduct e-commerce activities. A study by Gartner Inc. in December 2013 shows that tablets and smartphones account for around 86% of computing devices shipped in 2013. These mobile devices are also being increasingly targeted by malwares. A research by TrendMicro security lab indicates there are around one million variants of malwares detected in 2013 alone, targeting the Android platform. This shift of attacks to mobile devices bring in new challenges in operating system security that are not present in more traditional systems such as Windows and Linux, and require developments of new techniques to deal with the challenges. This project aims to develop runtime verification techniques, i.e., detection of attacks and anomalous behaviours at runtime, and to implement security monitors for mobile devices. More specifically, we aim to develop architecture independent and flexible security policy specification languages, and automated generation of security monitor codes to enforce policies specified in those languages. We have conducted a preliminary case study on the Android operating system, and shown that it is effective in mitigating some previously unknown attacks on Android systems.

Runtime verification is a process that consists of a number of activities: sensing of events (e.g., system calls, network packets sent/received, user login attempts), interpretation of events, and action specification and execution. Sensing of events can be done passively, e.g., observing network packets, or actively by instrumenting sensor codes in software or additional logic gates in hardware. The most important part of runtime verification is the interpretation of events. This is the part which determines whether a desired/undesired behavior is detected. This project focuses on the the development of a verification engine which can detect these kind of behaviors. The last part, action specification/ execution, deals with the actions that need to be performed upon detection of an undesired behavior.

As a major case study, we have designed and implemented a security extension of the Android OS with a custom security monitor. The Android OS is built on top of Linux kernel, so at the most basic level, it inherits most of the security architecture of Unix/Linux. The architecture of Android is given in Figure 1. Applications in Android, however, do not run directly on Linux. Rather they run inside a virtual machine, called Dalvik Virtual Machine (DVM), that are insulated from the rest of the system. This method of sandboxing an application is similar to the sandboxing implemented in Java virtual machine, but does not include certain security features, such as the Security Manager. This sandboxing and the Linux security features Android inherits address the traditional operating system security issues. However, most of the interesting issues arise on the level of applications, and a separate security mechanism is required to deal with those.

The Android application framework provides several API to access certain functionalities of the device, such as making phone calls, sending SMS, querying GPS location information, or unique device ID (such as the IMEI number), etc. Access to these functionalities is controlled via the permission mechanism of Android. For each functionality of interests, Android associates a unique permission required to access that functionality. For example, to be able to connect to the internet, an app in Android needs to have the INTERNET permission.

Similarly, to access fine location information (such as that obtained through a GPS device), an app needs to have the ACCESS FINE LOCATION permission. Permissions are explicitly requested by an app at installation time. But once they are granted, the app does not need to request them anymore every time it wants to access the functionalities associated with the permissions. Permissions granted to an app cannot be revoked without uninstalling the app. Despite all the security mechanisms mentioned above, Android is still vulnerable to various attacks. We shall focus mostly on attacks that target the permission mechanism at the application level. A main weakness in the Android permission mechanism lies in the socalled permission leakage problem. In Android, an app can provide a “service” to another app. Through this provision of services, an app can “leak” certain capabilities to an unprivileged app. For example, an app which has access to the ability to make phone call could act as a proxy for another app. This leads to the problem of privilege escalation, i.e., an app obtains a permission it was not granted by exploiting other apps.

NTU Figure 1. Android Architecture (source: Wikipedia).

At the core of runtime verification is a reference monitor that guards access to certain objects in the (operating) system that a system owner would like to protect from unauthorized uses. In the context of Android, these resources could be access to private data such as contact database, location information, or functionalities such as the ability to make phone calls, SMS or opening a connection to the internet. A runtime verification framework should allow one to specifiy unambigiously the security policy to be implemented, and an enforcement mechanism for that policy. In this work, we follow an established approach to runtime verification using linear temporal logic (LTL), which is commonly used in runtime verification. Linear temporal logic extends the (classical) propositional logic with temporal operators that allow one to specify temporal relations between events in a system, something which is crucial in detecting malware behaviours in Android. One of the advantages of specifying security policies in LTL is that policy enforcement becomes an instance of the model checking problem, which has been extensively studied in the literature and many results are already known and can be reused.

For our purpose, we need to extend LTL with two additional features. One is the addition of metric operators, which allow one to specify exact time interval between events. The other is the addition of recursive definitions, which allow one to specify recursive patterns of events.



Contents

One such pattern that we shall be looking at is the transitivity of system calls, i.e., we would like to be able to specify a call chain between two apps, possibly making use of other intermediate apps. Such call patterns are an ingredient in detecting privilege escalation attacks. The logic resulting from adding these two features to LTL is called RMTL (for ’recursive metric temporal logic’). This is as far as we know the first application of metric temporal logic in runtime verification of a complex operating system such as Android. Various security extensions of Android exist, but none has the flexibility and the expressivity that our framework provides.

We have implemented our modification to the Android OS, called LogicDroid. At the heart of LogicDroid is a reference monitor that enforces security policies specified in RMTL. The framework consists of three parts. The first part of the framework deals with detection of events. The second one deals with the enforcement of policies, and the third one deals with automated generation of monitors. The architecture of LogicDroid is given in Figure 2.

NTU Figure 2. LogicDroid’s architecture.

This is still a prototype and it is still being developed further to include more software sensors and performance improvements.

We have tested LogicDroid against a set of benchmarks, and recently against an exploit in Android, that was discovered after the development of LogicDroid. The exploit, which is cataloged in the Common Vulnerability Database CVE-2013-6272, allows any app to make phone calls covertly without having the permission to do so, thereby circumventing Android’s built-in security mechanism. LogicDroid is able to detect the exploit and stop the illegal access.

Although we develop our framework primarily in Android, the underlying technology is system independent. The resulting security monitors generated from our algorithms can be plugged into different systems, as long as the appropriate sensors of security-related events are implemented correctly. The fact that our monitor could detect a previously unknown exploit indicates the approach is valuable. Currently this is more of an academic project, but we intend to further refine our approach and explore further integration with state-of-the-art systems and devices to facilitate a wider adoption of our security framework.

Point of Contact

Alwen Fernanto Tiu
ATiu@ntu.edu.sg

Physiologically Based Modeling for Anatomical Structure

Problem & Motivation

Simulation platforms for biological systems are normally based on global dynamics or simplistic cellular automata, as a result, scientific hypothesis verification methods on these systems are not well supported. We seek to build a simulation platform that combines multi-agents and global dynamics as a means towards explaining complex behaviors so as to better assist in hypothesis verification. The synthesized models of biological phenomena will then be input data for extracting adaptive algorithms for constrained optimization.

Approach & Novelty

Researchers have proposed many methods on this topic that can be classified into two categories, geometrically based (or, non-physically based) and physically based methods. The non-physically based methods which require much less computation is mostly used for real-time applications, such as

games and virtual surgery. For example, shape matching method, position based method, and oriented particles can produce visually plausible results for virtual materials. However, they are not physically constrained thus the results are not accurate. The physically based method to which continuum mechanics is applied and commonly solved by the finite element method (FEM), can produce physically reliable results and be used to analyze the mechanical behaviors of real materials.

We propose the physically based simulation of elastic objects, for the purpose of analyzing the physical behaviors during elastic deformation, especially for the anisotropic soft tissues with specific internal structures, We have focused on modelling muscular tissues that have their own specific internal structures that determine the dynamic behaviors. We proposed a fiber orientation guided deformable model that can approximate the anisotropic elastic material properties. This proposed approach can provide a control of the desired deformation behavior and yield more physically plausible anisotropic behaviors for fiber constructed objects.




Contents

Significance & Impact

Early detection and diagnosis of the disease relies on both the image processing and computer vision techniques and the ability to quickly scan, retrieve and analyze information from relevant databases. Surgical treatment is then rendered more error resilient when assisted by augmented reality based visualization. Biomedical knowledge discovery and biological system simulation determines hidden patterns and correlations within various types and scales of medical data and thus assists in understanding and treatment of the diseases.

NTU Figure 2: A beating heart simulation with constraint points and muscular contraction

Point of Contact

LIN Feng
ASFLIN@ntu.edu.sg
COE-SCE Seed Funding

Multi-modal Feature Learning for Big Visual Data Analytics

Recently, the availability of large amount of labeled data has greatly boosted the development of feature learning methods for classification. In particular, convolutional neural networks (CNN) achieve great success in visual recognition/classification tasks. Features extracted from CNN can provide powerful global representations for the single object recognition problem. However, existing deep learning works typical only focus on single modality. NTU Where there exist multiple modalities data such as image and audio, image and captions, or RGB image and depth image, they simply extract features from each modality and then concatenate them together. Such a direct combination ignores the correlations and discriminations among different modalities. We have done a series of works on how to effectively and efficient fuse features from different modalities. We have also extended the existing single-object recognition task to multi-label or multi-object scenarios, which is more practical.

  • Motivation: how to make sense of billions of images & videos?
  • Problems: large-scale semantic object detection & recognition?
  • Approach: feature learning
  • Applications: image search, object detection & recognition, scene understanding

NTU

3D Visual Computing for Robotics, Healthcare
and AR Applications

3D technology has achieved great progress including the increasing popularity of 3D capturing devices such as laser scanner, depth sensor (e.g. Kinect), and the maturity of various image based 3D generation technology including multi-view stereo (MVS) technique, photometric 3D reconstruction and depth-based 3D reconstruction as well as the emerging 3D printing technology. We envision that when 3D media become easily reachable by common internet users in the way that is as simple as creating, playing, searching and delivering images & videos, an Internet 3D visual media era will come. We have done a series of works on 3D point cloud registration and reconstruction, 3D point cloud segmentation, labeling and understanding, aiming for related applications such as using 3D scene understanding and object recognition for robotics, 3D motion tracking for healthcare and 3D insertion for AR applications.

  • Motivation: More & more 3D information becomes available.
  • Problems: How to create 3D, track them and utilize them for better recognition?
  • Approach: use commodity depth cameras
  • Significance: bring 3D to common users

NTU

Point of Contact

CAI Jianfei
ASJFCai@ntu.edu.sg




Contents

Introduction

Over the last three decades, tremendous development has been observed in the field of Brain Computer Interface (BCI) technology. The primary focus of BCI technology is to develop a communication and control medium for severely disabled people with neuromuscular disorder. However, with the understanding of brain functionalities and the needs and potentials of the users, BCI systems have been developed for a vast range of applications including assistive and rehabilitative medical devices, entertainment technology etc. BCI consists of hardware and software modules that allow direct communication of brain with external environment. It involves detection and interpretation of the brain signals to translate user’s intentions to a computer or assistive technology devices, rehabilitation robots, prosthetic devices etc.

The primary focus area of BCI research is the neurological disorder that affects the motor cortex of the brain. This has resulted in the development of motor control-BCIs that function depending on the sensory motor rhythms (SMR) of brain. SMR-BCIs focus on characterizing and differentiating the neural features responsible for limb movement tasks. With the aid of signal processing and neural network tools, distinct event related patterns and SMR band powers for detecting movement task can be extracted translated into BCI output. This project investigates a lesser explored area of Electroencephalography (EEG) based BCI- Decoding movement kinematics of upper limb movement such as trajectory, speed, direction, etc. The challenge involves using non-invasive EEG that lacks sufficient spatial resolution to contain movement parameter information, to decode voluntary hand movement execution kinematics. The project reports dependent-BCI systems using EEG to analyze brain activity as the subject performs multi directional and multi speed center-out reaching hand movement tasks and to estimate the varying movement kinematics from these activity.

Research Objectives

The project aims to develop SMR-BCI systems using EEG to investigate neural correlates of hand movement kinematics. The specific objectives are as follows:

  • To design a voluntary hand movement execution experiment to validate the BCI system.
  • To develop signal processing techniques for investigating distinct brain activation patterns in spatial, spectral and temporal domains modulated by movement kinematics.
  • To evaluate performance of the BCI system to classify hand movement speed and direction using neural features derived from EEG.
  • To develop algorithms for continuous reconstruction of hand movement trajectory.

Highlights

The EEG based SMR-BCI system to investigate hand movement kinematics is validated using brain data collected during a voluntary hand movement task. The task involved during center-out hand movement reaching tasks in four orthogonal directions at two different speeds in a horizontal plane.

NTU Figure 1. BCI system: Block Diagram

The system block diagram is indicated in Figure 1. Signal processing algorithms to enhance signal quality and to extract optimal features from EEG data are developed. The reconstructed parameters are as indicated in Figure 2. The results obtained are significantly higher compared to the state-of-the-art methods.

The specific highlights of the project are listed below:

  • Hand movement speed: Signal processing algorithms that decompose multi-channel EEG signals in spectral and temporal domains are developed. SMR-BCI system that can differentiate binary speed movement tasks and can provide accurate classification and reconstruction of the recorded movement speed is developed using EEG.
  • Hand movement direction: EEG-based signal parameter that quantifies the presence of hand movement direction information is defined. Multi-class classification algorithm that determines the movement direction from the extracted neural features is developed.
  • Hand movement trajectory: Independent variables that quantify EEG activations in space, time and frequency domain are derived and used in linear regressor algorithm to model various parameters of hand movement kinematics. An adaptive estimation/reconstruction algorithm that determines the instantaneous position and speed of hand movement is also introduced.

NTU Figure 2. Recorded and estimated parameters.




Contents

Applications and Future Scope

The major goals of this research were to develop EEG-BCI systems specifically focusing on defined and higher dimensional motor control output. The results indicate the potential of EEG-BCI system in motor control applications such as wheelchair, robotic limbs, quad copters/drones etc., to perform a movement at user-intended parameters.

Further, the system can be modified to reconstruct imagined movement paths in a real time scenario. Such a system will have

great potential in neurorehabilitation and neuroprostheses as a neurofeedback movement training system for disabled users. The ultimate goal is to develop a realistic, efficient and robust SMR-BCI system that can contribute to real-world applications.

Point of Contact

Vinod A PRASAD
ASVinod@ntu.edu.sg
Neethu ROBINSON

Modeling, Validation, and Analysis of Gene Regulatory Networks
including Delays

Summary

Gene regulatory networks (GRNs) are to cells as electrical circuitries are for devices. A GRN consists of nodes (each representing a gene or its products, i.e. RNA or protein) and links between nodes (each representing biochemical interaction between two molecules). Knowing the wiring of circuits in an electronic device allows it to be fixed when it does not work. Likewise, knowing the wiring of biological networks in cells help find effective therapies to human diseases. However, network structures inside cells are largely unknown. Therefore, in order to unveil the hidden mechanisms of diseases, it is highly desirable to map the regulatory networks controlling the dynamics of cells. The strategy of systems biology is to reverse engineer the architecture of living systems from high-throughput data, which are snapshots of the dynamic systems. One prevalent type of data is gene expression data, which consists of measured abundances of products of genes (mostly mRNA, but sometimes proteins) along a series of time points. In our project, a major goal was to reconstruct GRNs from time-series gene expression data. Since the regulatory events take different amounts of time, time-delay is a critical property of a GRN system.

The inference of GRN from time series gene expression data is faced with many challenges. One major challenge is that, as the number of time points tends to be smaller than the number of genes, the GRN inference is often instable, in the sense that different GRN structures can explain the same dataset equally well. We proposed statistical methods to handle this issue. One method is to introduce random perturbations in the gene expression data, and identify the network structures in convergence as the top candidates. Another method is to integrate additional data (e.g. protein-protein interactions, epigenetic profiling). However, both of our theoretical and simulation studies suggested that increasing time points would be needed to ensure stability of GRN inference. Moreover, we developed new methods to estimate time delays from time-series gene expression data, and observed that including time delays can improve the accuracy of GRN inference. Another challenge in the field is how to validate predicted GRN, since in most cases benchmark networks are unavailable. To address this issue, we developed text-mining methods to extract evidence of time delays and gene regulations from biological literature. After obtaining reliable GRN, we analyze the network topologies using statistical analysis, to identify core genes or network motifs (small sub-networks with high frequencies or importance). Applying the computational techniques to real biological experiments, we ........................

NTU


NTU Figure 1. Screenshots of the MetaGRN software developed in this project, for integrated inference and analysis of gene regulatory networks, with friendly user interface and network visualization.

identified two crucial signaling pathways (a kind of network motifs) behind the virus infection of H3N2, and aberrant immune responses of mouse (which is to model human responses). We have also applied network analysis to study of human diseases (e.g. cancer, malaria). Such systems-level analysis of network structures can provide mechanistic insight into biomedical questions unavailable through traditional wet-lab experiments. To facilitate fellow researchers, especially biologists without much training in computation, we developed a software tool called “MetaGRN” (Figure 1), which integrates different algorithms of GRN inference, with network visualization and intelligent choices of parameters.



Contents

Most existing algorithms of GRN inference are suitable only for small networks, of no more than 50 nodes. To scale to larger networks, we implemented two algorithms with high-performance computing techniques of parallelism, which significantly reduced the running time.

In this project, we have developed computational techniques to infer GRN with time delay, as well as validation and analysis of networks in order to answer biological questions. The outcomes of this project can greatly facilitate the research of wet-lab biologists as well as computational biologists. The network structures and parameters inferred with our methods can provide

a platform for biological knowledge discoveries and pharmaceutical developments (e.g. generating new hypotheses to be tested by wet-lab experiments, predicting effects of new drugs or therapeutic strategies).

Point of Contact

ZHENG Jie
ZhengJie@ntu.edu.sg

Multi-agent Simulation for Medical Nanorobotics

Nanorobotics, which is a new field in robotics technology, may someday benefit clinical medicine by delivering both drugs and diagnostics into the human body. Potential applications of medical nanorobotics include early diagnosis of cancer, neutralisation of viruses, precise and incisionless surgery, and monitoring and treatment of diabetes. In particular, targeted drug delivery (TDD), which uses nanoparticles functionalised with targeting moieties such as antibodies and DNA aptamers, bears the most promise and research interest in nanomedicine. Today, harmful side effects of cancer treatments such as chemotherapy are commonly a result of drug delivery methods that do not pinpoint their intended target cells accurately. Nanorobots, on the other hand, can be applied to chemotherapy to combat cancer through precise chemical dosage administration. The killer agents would be able to identify, target, and finally destroy solid tumors or atherosclerotic plaques by directing drugs and diagnostic molecules to sites of interest. A similar approach could also be taken to enable agents to deliver anti-virus drugs.

The design of nanorobotic systems requires the use of information from a vast variety of sciences. As the field of nanotechnology is still considered rather new, a lot of experiments and research are still needed to be done on the issues regarding the implementation of the medical nanorobots. Since nanoscale devices have not yet been fabricated, evaluations of possible designs, control algorithms, and potential applications using computer simulations and virtual environments are required. However, the need of proper modelling and simulation framework for medical nanorobotics is still largely unmet.

To study the application of nanorobotics in medicine, this project focuses on the design of a multi-agent system (MAS), which is a physicsbased computer simulation composed of multiple interacting intelligent agents within an environment. MAS has been used to solve problems that are difficult for an individual agent to solve. A swarm of simulated agents that has individual skills and partial information about a problem can work collaboratively to develop a solution to it. Typically, the MAS models have following general properties: agents are autonomous and independent; ............

agents are aware only of their local environment; in other words, no agent has all the data in the system; there is no central control component, therefore no agent would be able to supervise the whole system. In addition, to execute their tasks in such a system, agents usually need to have means of communication and mutual interactions. These MAS properties correspond clearly with the properties of the nanorobotic system that could be located in a human body. In such conditions, nanorobots need to be autonomous as there is no possibility for all of them to communicate; nanorobots are unable to ‘know’ the whole environment as they are constrained in terms of computational power and storage capabilities, there is no possibility to introduce one central component that would be able to supervise all nanorobots. Moreover, the system is completely asynchronous. It also needs to adapt to constantly changing properties in terms of agents’ spatial positions, their communication and interaction partners. In order to develop these properties, the agent-based system should include additional models such as chemical sensor, power supply, data communication, and collision detection capabilities. These models are used to realistically simulate the behaviours of the nanorobots and their environment.

This cross-disciplinary research lies at the intersection of mechanical and biomedical engineering, clinical medicine, artificial intelligence, and virtual reality. MAS is an essential tool for exploring the human body, giving a better understanding on the requirements, constraints, and working principles of nanorobots through the use of virtual reality technologies, and evaluating the strength and the usefulness of the robots. Some typical application scenarios such as TDD are being developed and studied, which will then provide a stepping stone for more research into nanomedical applications.

Point of Contact

Elvis LIU Sze Yeung
elvisliu@ntu.edu.sg

Optimal Offloading Algorithm for Mobile Cloud Computing with Cloudlet

With the fast development of mobile devices, cloud computing has been extended to a mobile paradigm. Devices, namely cloudlets, which are rich in computing resources, can provide cloud-like services to local mobile users via wireless connections. With such local cloud services supported by cloudlets, mobile users no

longer need to always access business level cloud providers such as Amazon. Instead, the requests for cloud resources can be directed to cloudlets to save energy, reduce latency and cut cost for cloud services. However, the cloudlet architecture gives rise to one critical issue that the underlying structure of the ............



Contents

cloudlet system is always changing due to the mobility feature. The number of accessible cloudlets and the connections from the mobile user to cloudlets may vary. The policy of the mobile user to use the cloudlet resources is not as easy as the pay-as-you-go policy as in regular cloud services.

To find an optimal policy to employ local mobile cloudlets under the highly varying and uncertain circumstance, the Markov decision process (MDP) model is applied to describe how the system gradually changes when cloudlets and mobile users are on the move. By solving the MDP model, the mobile user can find an optimal policy comparing to some baseline schemes (e.g., the conventional greedy scheme and random scheme). The MDP model takes future states of the mobile user into consideration, which makes the mobile user become smarter and have a more broaden view of the whole operation process. Moreover, this research has discovered a monotonic structure of the solved optimal policy, which is potentially important for finding fast and sub-optimal algorithms to decide the mobile user’s policy. The research may find its applications in the projects where mobile devices and sensors are mainly involved, such as the current Smart Nation project. The computing power of resourcepoor mobile devices (e.g., vehicular sensors, mobile phones, and smart meters) can be optimally enhanced with the assistance of the existing high computing power devices nearby, such as servers and base stations deployed by public sectors as a part of Smart

Nation architecture. The computing power of those resource-rich devices can also be efficiently used when they are idle.

NTU

Point of Contact

Dusit NIYATO
DNIYATO@ntu.edu.sg

Sentic Computing

Sentic computing is a multi-disciplinary approach to natural language processing and understanding at the crossroads between affective computing, information extraction, and common-sense reasoning, which exploits both computer and human sciences to better interpret and process social information on the Web. In sentic computing, whose term derives from the Latin ‘sentire’ (root of words such as sentiment and sentience) and ‘sensus’ (as in common-sense), the analysis of natural language is based on linguistics and common-sense reasoning tools, which enable the analysis of text not only at document-, page- or paragraph-level, but also at sentence-, clause-, and concept-level. In particular, sentic computing’s novelty gravitates around three key shifts:

  • Shift from mono- to multi-disciplinarity – evidenced by the concomitant use of AI and Semantic Web techniques, for knowledge representation and inference; mathematics, for carrying out tasks such as graph mining and multi-dimensionality reduction; linguistics, for discourse analysis and pragmatics; psychology, for cognitive and affective modeling; sociology, for understanding social network dynamics and social influence; finally ethics, for understanding related issues about the nature of mind and the creation of emotional machines.

NTU

NTU

  • Shift from syntax to semantics – enabled by the adoption of the bag-of-concepts model in stead of simply counting word co-occurrence frequencies in text. Working at concept-level entails preserving the meaning carried by multi-word expressions such as cloud_computing, which represent ‘semantic atoms’ that should never be broken down into single words. In the bag-of-words model, for example, the concept cloud_computing would be split into computing and cloud, which may wrongly activate concepts related to the weather and, hence, compromise categorization accuracy.
  • Shift from statistics to linguistics – implemented by allowing sentiments to flow from concept to concept based on the dependency relation between clauses. The sentence “iPhone6 is expensive but nice”, for example, is equal to “iPhone6 is nice but expensive” from a bag-of-words perspective. However, the two sentences bear opposite polarity: the former is positive as the user seems to be willing to make the effort to buy the product despite its high price, the latter is negative as the user complains about the price of iPhone6 although he/she likes it.

Contents

The core element of sentic computing is SenticNet, a knowledge base of 50,000 common-sense concepts. Unlike many other sentiment analysis resources, SenticNet is not built by manually labelling pieces of knowledge coming from general NLP resources such as WordNet or DBPedia. Instead, it is automatically constructed by applying graph-mining and multidimensional scaling techniques on the affective common-sense knowledge collected from three different sources, namely: WordNet-Affect, Open Mind Common Sense and GECKA. This knowledge is represented redundantly at three levels: semantic network, matrix, and vector space. Subsequently, semantics and sentics are calculated though the ensemble application ofspreading activation, neural networks and an emotion categorization model.

SenticNet can be used for different sentiment analysis tasks including polarity detection, which is perfomed by means of sentic patterns. In particular, a semantic parser is firstly used to deconstruct natural language text into concepts. Secondly, linguistic patterns are used in concomitance with SenticNet to infer polarity from sentences. If no match is found in SenticNet or in the linguistic patterns, machine learning is used.

NTU

Point of Contact

Erik CAMBRIA
cambria@ntu.edu.sg
With permission from sentic.net/computing




Contents
Project Description Agency Principle Investigator
A Collaborative Computing Approach to Consistency Maintenance in Cloud Storage Ministry of Education, Singapore SUN CHENGZHENG
Securify: A Compositional Approach of Building Security Verified System Ministry of Education, Singapore CONG GAO
Affordable U-healthcare Platform for Diabetes Management National Healthcare Group ARVIND EASWARAN
Anomaly Understanding: Visual Attention Modelling for Vision-related Tasks ST Dynamics Pte Ltd LIN WEISI
Biharmonic B-splines on Manifold Triangle Meshes Ministry of Education, Singapore HE YING
BOSSANOVA DSO National Laboratories ERIK CAMBRIA
Classification & Characterization of Hyperpigmented Spot types Agency for Science, Technology and Research (A*STAR) KONG WAI KIN ADAMS
Cognitive team theoretic approach for Dynamic Airspace Management (CDAM) Civil Aviation Authority of Singapore SURESH SUNDARAM
Collecting and Analyzing Data from Smart Device Users with Strong Privacy Protection Samsung Electronics Co., Ltd. XIAO XIAOKUI
Computational Optical Precision Measurement: Local or Non-local? Ministry of Education, Singapore QIAN KEMAO
Computational Virulence Model with Functional Information for Influenza Viruses Ministry of Education, Singapore KWOH CHEE KEONG
Crowd Sourcing for Robust and Efficient Scene Perception and Assessment of Urban Mobility National Research Foundation SRIKANTHAN THAMBIPILLAI
Crowd-Scale, Data-Driven Assessments of Coding Ability in Large Computing Courses Ministry of Education, Singapore XING ZHENCHANG
Crowdsourcing a Personalized Practice Environment for Mathematics through Data Analytics Ministry of Education, Singapore HUI SIU CHEUNG /
SINNO JIALIN PAN
C-RT3.3: Automated Flow Feature Detection and Tracking on Unsteady CFD Simulation Data Rolls Royce Corporate Lab KE YIPING, KELLY,
XIAO XIAOKUI
Cyber Security Solutions For Smart Traffic Control Systems National Research Foundation BO AN
Data Driven Human Preference Analytics for Urban Computing Microsoft Operations Pte Ltd CONG GAO
Determinantal Point Process for Modelling RF Energy Harvesting Networks Ministry of Education, Singapore DUSIT NIYATO
Development Of A Decision Support Tool Using Modeling For Defining The Optimal Production And Distribution Solutions For Cold System In Urban Area Veolia City Modelling Center Pte Ltd-NTU Joint R&D ARVIND EASWARAN
Full-Duplex Transceivers for 5G Wireless Systems Ministry of Education, Singapore A S MADHUKUMAR
Haptic Interaction with Images and Videos Ministry of Education, Singapore ALEXEI SOURIN
Hybrid Finite Element Method and Mixedlevel Coarse Graining Molecular Dynamics simulation Ministry of Education, Singapore KWOH CHEE KEONG
Intelligent CAD Algorithms using Cloud Computing and Machine Learning Ministry of Education, Singapore NACHIKET GANESH KAPRE
Interference Management for Emerging Satellite Communication Networks Airbus Group Singapore Pte Ltd A S MADHUKUMAR
Isogeometric Design and Simulation Using T-NURCC Ministry of Education, Singapore ZHENG JIANMIN
Logic Synthesis Flow for Digital Design Technologies Beyond CMOS Ministry of Education, Singapore ANUPAM CHATTOPADHYAY
Mathematical Modeling of Cyber Threats Using Real-time Finite-State Automata Ministry of Education and Science of the Russian Federation NG WEE KEONG



Contents
Project Description Agency Principle Investigator
Mechanized Equivalence Proofs for Security Protocols Ministry of Education, Singapore ALWEN FERNANTO TIU
NTU-AIA Collaboration Projects (Edge Lab) AIA Company Limited SRIKANTHAN THAMBIPILLAI
NTU-PKU Research Institute Research Project Ng Teng Fong Charitable Foundation Limited MIAO CHUN YAN
Parameter Estimation for Convolutional Codes in the Presence of Bit Errors Ministry of Defence, Singapore A S MADHUKUMAR
Peer-to-Peer Synchronization: Cognitive Team Theoretic Approach for Machine Collaboration ST ENGINEERING-NTU Joint R&D SURESH SUNDARAM
Properties and Interactions of Functional Modules of the Human Brain across the Life Span Ministry of Education, Singapore JAGATH C RAJAPAKSE
Research and Development in Mobile Phone Security - III MINDEF-NTU Joint R&D NG WEE KEONG
Research on Integrated Analysis for Cyber-Physical Security NEC Asia Pacific Pte Ltd NG WEE KEONG
Security Enhancements For ATM And POS Systems Wincor Nixdorf Pte Ltd LIU YANG
Semantic Matching and Sense Making: Understanding What People Saying on the Web Ministry of Education, Singapore SUN AIXIN
Software Vulnerability Discovery Tool Building DSO National Lab-NTU Joint R&D LIU YANG
Spectral Efficiency Technologies for Aeronautical Communication Systems Airbus Group Singapore Pte Ltd A S MADHUKUMAR
SUMO: Securing Untrusted Memories in Embedded Systems Ministry of Education, Singapore LAM SIEW KEI
Think on Your Feet - Learning through Purposeful Collaborative Play and Movement Singapore Millenium Foundation GOH WOOI BOON
TOD: Towards Truly On-Demand Resources for Latency-sensitive Services on Public Clouds Ministry of Education, Singapore TA NGUYEN BINH DUONG
Towards Data-driven Visual Graph Query Interface Construction and Management Ministry of Education, Singapore SOURAV SAHA BHOWMICK
Ultra Low-Power Embedded Machine Learner For Image/Video Processing In Robotics And Portable/ Wearable Devices Singapore-MIT Alliance for Research & Technology (SMART) NACHIKET GANESH KAPRE
Unobtrusive Sensing for Aging-In-Place National Research Foundation MIAO CHUN YAN
Validation of an Algorithm that Automatically Segment and Calculate the Area affected by Vitiligo on the Trunk of Patients for Assessment of Extent of Disease and Response to Treatment in Patients with Vitiligo National Skin Centre, Singapore KONG WAI KIN ADAMS
Whom-To-Follow Recommender Systems in Interest-Social Networks: A Fashion Case Study TELENOR Group ZHANG JIE
Contents
Project Description Agency Principle Investigator
A Cooperative Co-Evolutionary Framework for Data-driven Model Calibration Ministry of Education, Singapore CAI WENTONG
A High Performance Stowage Planning and Optimisation System for Large Container Vessels Neptune Orient Lines Ltd HSU WEN JING
A Study of Online Social Forums for Enabling Community Driven Healthcare Support Ministry of Education, Singapore SUN AIXIN
Altmetrics: Rethinking and Exploring New Ways of Measuring Research Outputs National Research Foundation XIAO XIAOKUI
Anonymization of Complex Data with Strong Privacy Guarantees Ministry of Education, Singapore XIAO XIAOKUI
Array-based Database Technology for Large-scale Satellite Data and their Analysis Ministry of Education, Singapore HO SHEN-SHYANG
CACHE ME IF YOU CAN: Optimizing On-chip Memory Usage for Multi and Manycore Processors Ministry of Education, Singapore FAN RUI
Computational Cognitive Big Data Analytics System Ministry of Education, Singapore SURESH SUNDARAM
C-RT1.1: Robust Large Vocabulary Continous Speech Recognition (LVCSR) for Far Field recordings Rolls Royce Corporate Lab CHNG ENG SIONG,
ERIK CAMBRIA
C-RT1.2: Ontology-Based Text Mining from Speech Data Rolls Royce Corporate Lab ERIK CAMBRIA,
CHNG ENG SIONG
C-RT2.1: Image Data Knowledge Elicitation Rolls Royce Corporate Lab HO SHEN-SHYANG,
CHIA LIANG TIEN
CLEMENT
C-RT2.2: Context-Aware Exploration of Image Data Rolls Royce Corporate Lab HO SHEN-SHYANG,
CHIA LIANG TIEN
CLEMENT
C-RT3.1: Virtual Engine Emulator by Using Data Fusion Rolls-Royce Corporate Lab ONG YEW SOON,
KONG WAI KIN ADAMS,
KE YIPING, KELLY
C-RT5.1: Framework for Rapid Simulation of Complex Business Systems Rolls-Royce Corporate Lab ONG YEW SOON,
ZHANG JIE
C-RT5.2: Business Process Simulation Description Language Rolls-Royce Corporate Lab XING ZHENCHANG,
LIU YANG
Data Distribution for Exascale Distributed Virtual Environment Simulations Ministry of Education, Singapore LIU SZE YEUNG,
ELVIS
Distributed Plant Modelling, Fault Diagnosis, and Supervisor Control of Large Scagle Automated Manufacturing Systems Ministry of Education, Singapore LIU YANG
Electroencephalogram-based Hybrid Brain-Computer Interface System for Biometric Identification Ministry of Education, Singapore VINOD ACHUTAVARRIER PRASAD
iClonecast-cloud-based service engine for multi-screen applications Singapore -MIT Alliance for Research & Technology (SMART) WEN YONGGANG
Impact of Diversity and Module Level Storage for Mitigating against Solar PV Variability in Singapore Ministry of Education, Singapore DOUGLAS MASKELL
Improving Reliability and Economic Performance of Microgrids Economic Development Board, Singapore WANG PING
Investigation into Perception-inspired Mobile Graphics (PMG) Ministry of Education, Singapore LIN WEISI
Knowledge Discovery from Mobile Communication Networks and Its Applications Ministry of Education, Singapore KE YIPING, KELLY
MAISON: Minor Languages Conversational Speech Recognition DSO National Laboratories CHNG ENG SIONG
Mixed-Criticality Scheduling for Cyber-Physical Systems Ministry of Education, Singapore ARVIND EASWARAN
Multi-agent Evolutionary Framework for Multi-objective Vehicle Routing Ministry of Education, Singapore ZHANG JIE
Optimal Resource Allocation in Multi-Radio Multi-Channel Wireless Networks Ministry of Education, Singapore WANG PING



Contents
Project Description Agency Principle Investigator
PADDINGTON MINDEF-NTU Joint R&D ERIK CAMBRIA
Physiologically-Based Modelling and Visualization of a Beating Heart Ministry of Education, Singapore LIN FENG
Privacy-Preserving Database Systems on Untrusted Servers Ministry of Education, Singapore NG WEE KEONG
Real-time Indexing and Search of Large logs in Dynamic Environments HUAWEI Technologies Co. Ltd XIAO XIAOKUI
Saddle Vertex Graph (SVG): A Novel Solution to the Discrete Geodesic Problem Ministry of Education, Singapore HE YING
Securify: A Compositional Approach of Building Security Verified System National Research Foundation SRIKANTHAN THAMBIPILLAI,
LIU YANG,
ALWEN FERNANTO TIU
Semantic 3D labeling & reconstruction with RGB-D data Ministry of Education, Singapore CAI JIANFEI
Social Computational Analytics for Trend Discovery and Social Media Marketing Ministry of Education, Singapore ERIK CAMBRIA
Synthetic Data Generation with Differential Privacy Advance Technologies and Architecture (AT&T) XIAO XIAOKUI
Theoretical Study and Technical Design of Green Memory Cloud Systems for Real-time Data Analytics Ministry of Education, Singapore HE BINGSHENG
Towards a Big-data Platform for Social TV Analytics Ministry of Education, Singapore WEN YONGGANG
Towards Design and Modelling of Brain-Inspired Cognitive Systems Ministry of Education, Singapore TAN AH HWEE
Towards Scalable Distributed Cloud-Based MMOG Services Ministry of Education, Singapore TANG XUEYAN
Using Body Geometry as Supportive Information for Personal Identification Ministry of Education, Singapore KONG WAI KIN ADAMS
Vulnerability Detection in Binary Code Centre for Strategic Infocomm Technologies LIU YANG



Contents
Name of Candidate Research Topic Supervisor
Alok Prakash Constraint-Aware Configurable System-on-chip Design for Embedded Computing Thambipillai Srikanthan
Ang Hock Hee Distributed Classification in P2P Networks Ng Wee Keong
Arfika Nurhudatiana Relatively Permanent Pigmented or Vascular Skin Marks (RPPVSM) for Forensic Identification Kong Wai-Kin Adams
Cao Xin Spatial Keyword Querying Beyond the Single Geo-Textual Object Granularity Cong Gao
Chen Lin Advanced Maximum Margin Learning Techniques for Large Scale Vision Tasks Xu Dong
Chiew Wei Ming Architectural and Algorithmic Design for Embedded Medical Imaging Lin Feng
CHUA HUEY ENG TRECENTO: In Silico Network-Driven Identification of Target Combinations for Combination Therapy Sourav S Bhowmick
Fang Hui Coping with Subjectivity and Dishonesty in Opinion Evaluation by Exploiting Social Factors Zhang Jie
Feng Liang A Study on Probabilistic Memetic Algorithm and Meme-Centric Framework Ong Yew Soon
Feng Shu Computational Techniques for Modeling Non-player Characters in Games Tan Ah Hwee
Giduthuri Sateesh Babu Meta-Cognitive Sequential Learning In RBF Network for Diagnosis of Neurodegenerative Diseases Suresh Sundaram
Gu Yuan Emotion Recognition: From Physiological Signals to Affective States Ng Wee Keong
Guo Guibing Exploiting Ratings and Trust to Resolve The Data Sparsity and Cold Start of Recommender Systems Zhang Jie
Hai Zhen Aspect-based Opinion Mining of Customer Reviews Cong Gao
Jiang Lianlian Modeling and Optimization of Photovoltaic Systems under Partially Shaded and Rapidly Changing Conditions Douglas Leslie Maskell
Jiang Siwei Multiagent Evolutionary Computation for Complex Problems Zhang Jie
Jin Jiangming HPC Simulations of Information Propagation over Complex Networks Stephen John Turner
Kang Yilin Computational Methods for Modeling Human-like Agents in Virtual Environment Tan Ah Hwee
Karthik Muthuswamy Algorithms for Saliency Detection in Videos Deepu Rajan
Kartick Subramanian Meta-Cognitive Learning Algorithm for Neuro-Fuzzy Inference Systems Suresh Sundaram
Lai Danbo Interactive Shape Modelling for Mathematics Visualization Alexei Sourin
Le Thanh Hai Scalar & Homoskedastic Models for SAR & POLSAR Data Vun Chan Hua, Nicholas
Lee Zhuo Qi New Results on Spread of Influence in Social Networks Hsu Wen Jing
Li Feng Designing Localized Algorithms for Large-scale Wireless Sensor Networks: A Geometric Perspective Luo Jun
Li Wen Visual Recognition by Learning from Web Data Xu Dong
Lin Miao Uncovering Individual's Mobility Patterns from GPS Dataset Hsu Wen Jing
Lin Wenqing Efficient Techniques for Subgraph Mining and Query Processing Xiao Xiaokui
Liu Cheng Collaborative Interest Management for Peer-to-Peer Massively Multiplayer Online Games Cai Wentong



Contents
Name of Candidate Research Topic Supervisor
Liu Jigang Conditional Bayesian Filtering for Robot Navigation and Human Tracking Deepu Rajan
Liu Yuan Towards the Design of Robust Incentive Mechanisms to Address Subjectivity and Dishonesty Problems in Agent Reporting Zhang Jie
Lua Rui Ping Anti-fragile Internet with Autonomous Swarm Networks Ng Wee Keong
Meng Lei Clustering and Heterogeneous Information Fusion for Social Media Theme Discovery and Associative Mining Tan Ah Hwee
Neethu Robinson Signal Processing Techniques for Motor Control Brain Computer Interface Systems Vinod Achutavarrier Prasad
Nguyen Trung Hieu Speaker Diarization in Meetings Domain Chng Eng Siong
Patanaik Amiya Vulnerability to Sleep Deprivation: A Drift Diffusion Model Perspective Kwoh Chee Keong
Ponnu Jacob Harmonized Co-existence with Robust Interference Management in Femto-Macro Cellular Architectures A S Madhukumar
Rajendra Prasad Sirigina Efficient Relaying Strategies for Mitigating Fading and Interference A S Madhukumar
Renuga Kanagavelu Software Defined Networks and Virtual Machine Placement for Data Center Networks Lee Bu Sung
Shahzad Rasool Tangible Images Alexei Sourin
Sinha Sharad Intelligent High Level Synthesis for Customization on Reconfigurable Platforms Thambipillai Srikanthan
Su Tran To Chinh Improving the Discrimination of Near-Native Complexes for Protein Rigid Docking by Implementing Interfacial Water into Protein Interfaces Kwoh Chee Keong
Sudarshan Guruacharya Radio Resource Management in Heterogeneous Wireless Networks Dusit Niyato
Sun Hequan Effective Graph-based Algorithms for Weak Motif Discovery in Genomic Sequences Hsu Wen Jing
Tan Mingkui Towards Efficient Large-scale Learning by Exploiting Sparsity Ong Yew Soon
Tan Ngan Meng Medical Imaging Algorithm Research for Diagnosis of Ocular Diseases Goh Wooi Boon
Tan Wi-Meng Javan A Self-reorganizing Neuro-fuzzy Associative Machine for Algorithmic Financial Time-series Modeling Quek Hiok Chai
Tang Peng Handwriting Recognition and Retrieval for Chemical Structural Formulas Hui Siu Cheung
Tang Shanjiang Optimization Techniques on Job Scheduling and Resource Allocation for Mapreduce System Lee Bu Sung
Thaw Tar Thein Zan Algorithms For Detection and Localization of Events in a Water Distribution Network Lee Bu Sung
Thidathip Wongsurawat Computational Analysis and Prediction of Specific Genomic Regions Forming R-loop Structure and Chromosomal Variations Associated with Cancer Vladimir Kuznetsov
Tran Hoang Ha In Network Processing for Tectonic Monitoring Lee Bu Sung
Vaisagh Viswanathan Thattamparambil Modelling Behavior in Agent Based Simulations of Crowds Bo An
Vipin Kizheppatt Design Automation for Partially Reconfigurable Adaptive Systems Suhaib A Fahmy



Contents
Name of Candidate Research Topic Supervisor
Vishram Mishra A QoS Provisioning Framework for Cognitive Radio Network Lau Chiew Tong
Vivek Balachandran Software Protection through Obfuscation Ng Wee Keong
Wang Wenjie Solving Multi-Agent Path Planning by Local Search Algorithms Goh Wooi Boon
Wang Wenwen Neural Modeling of Multiple Memory Systems and Learning Tan Ah Hwee
Wu Qiong Modeling Curiosity for Intelligent Agents Miao Chun Yan
Wu Xiaoqun Variational Methods for Geometry Processing Zheng Jianmin
Wu Zhizheng Spectral Mapping for Voice Conversion Chng Eng Siong
Xia Hao Learning with Multiple Kernels: Algorithms and Applications Hoi Chu Hong
XU SHUOYU Computer Aided Diagnosis of Liver Fibrosis using Non-linear Optics Microscopy Sourav S Bhowmick
Xu Xinxing Learning with Multiple Representations: Algorithms and Applications Xu Dong
Yang Ming Advanced Scheduling in Data Transmission and Mobile Media Cloud Computing Cai Jianfei
Yuan Quan Exploiting Spatial, Temporal, and Semantic Information for Point-of-interest Recommendation Cong Gao
Zhang Chi Towards A Large Scale Indoor Localization Service with Crowdsensing Indoor Map Generation Luo Jun
Zhang Hengyi Vein Identification for Forensic Investigation Kong Wai-Kin Adams
Zhang Minqi Constrained Conformal/Quasi-Conformal Map and Its Applications He Ying
Zhang Tianyou Contact Network Based Framework for Infectious Disease Interventions Kwoh Chee Keong
Zhang Weiwen Towards Energy Efficient Task Execution in Mobile Cloud Computing Wen Yonggang
Zhang Yang Stochastic Optimizations of Mobile Energy Management Dusit Niyato
Zhang Yu Efficient Representation for Image Recognition Cai Jianfei
Zhang Yuzhe Techniques for Body and Garment Modeling Zheng Jianmin
Zhang Zhuo Biomedical Imaging Informatics in Ocular Disease Diagnosis Kwoh Chee Keong
Zheng Hanying Server Provisioning for Distributed interactive Applications Tang Xueyan
Zheng Yuanqing Fundamental Study on Improving RFID System Operational Efficiency Li Mo
Zhong Haoming Temporal Fuzzy Cognitive Maps for Corporate Credit Ratings Miao Chun Yan
Zhu Diwen Efficient Query Processing on Large Graphs Xiao Xiaokui
Zhu Hongyuan Semantic Image Segmentation & Cosegmentation Cai Jianfei
ZHU SHIWEN Quantification of Subcellular Localization and Co-localization from High-content Images Kwoh Chee Keong
Contents
Name of Candidate Research Topic Supervisor
Ailiya Affective Teachable Agent in Virtual Learning Environment Miao Chun Yan
Amrith Dhananjayan V M A Designer Support Framework for Specification Comprehensibility in Automata-based Discrete-Even Control Design Seow Kiam Tian
Arun Kumar Energy-Efficient Broadcasting in Wireless Ad-hoc Sensor Networks Lee Bu Sung
Arun Kumar Time Frequency Analysis of Functional MR Images Lin Feng
Behrouz Saghafi Khadem Human Action Recognition by Embedding Silhouettes and Visual Words Deepu Rajan
Chen Xianshun A Study on Memetic Computation, with Applications to Capacitated Vehicle Routing Problems Ong Yew Soon
Chen Yupeng Design and Analysis of Bioinfomatics Algorithms on an FPGA Platform Douglas Leslie Maskell
Cui Jin High Level Thermal-aware Scheduling for Multiprocessors Douglas Leslie Maskell
Darak Sumit Jagdish Design of Low Complexity Variable Digital Filters and Reconfigurable Filter Banks for Multi-standard Wireless Communication Receivers Vinod Achutavarrier Prasad
Dong Qiumin Reliability Analysis and Performance Optimization of Wireless Networks and Its Applications Dusit Niyato
Duan Qi Lighting Geometry Aware Environment Matting and 3D Reconstruction Cai Jianfei
Ehsan Shahrian Varnousfaderani Sampling-based Approaches for Image and Video Matting Deepu Rajan
Ehsan Younessian A Framework for Associated News Story Retrieval Deepu Rajan
Elahe Farahzadeh Tools for Visual Scene Recognition using the Local Approach Cham Tat Jen
Erwin Anggadjaja Cross-layer Mimo-links Exploiting Packet Switching and Adaptive Modulation for TCP/IP Enabled Volcano Monitoring Networks Lau Chiew Tong
Fan Hongfei Any-Time Collaborative Programming Environment and Supporting Techniques Sun Chengzheng
Fang Shuo Data Traffic Congestion Management for Data Center Ethernet Chang Kuiyu
Goh Weihan Towards Inconspicuous Security Yeo Chai Kiat
Gong Shimin Robust Optimization and Its Applications in Cognitive Radio Networks Wang Ping
Gu Zhouye Advanced Video Coding Based on Matrix Decomposition Lin Weisi
Han Xiaogang Remembrance Agent: from Theory to Applications Miao Chun Yan
Hari Vishnu Improved Techniques for Detection and Localization of Underwater Acoustic Sources using acoustic Vector Snesors Annamalai Benjamin Premkumar
Hu Nan Spatio-Temporal Patterns and Pedestrian Simulation Cai Wentong
Iti Chaturvedi Bayesian Framework for Building Gene Regulatory Networks Including Delays Kwoh Chee Keong
Jalaluddin Qureshi Achieving Reliability for Wireless Multicast Transmission Using Networks Coding Cai Jianfei
Jonathan William Dennis Sound Event Recognition in Unstructured Environments using Spectrogram Image Processing Chng Eng Siong
Li Bin On-line Portfolio Selection Hoi Chu Hong
Li Chenliang Harnessing Online Social Media to Deal with Information Overload Anwitaman Datta



Contents
Name of Candidate Research Topic Supervisor
Li Guangxia Collaborative Learning From Multiple Data Sources Chang Kuiyu
Li Yifan Game-Theoretic Modeling for Resource Allocation in Relay-based Wireless Networks Wang Ping
Li Yusen Consistency-Aware Scheduling and Load Balancing in Multi-server Distributed Virtual Environments Cai Wentong
Lu Zongqing Exploring Social Structures for Network Protocol Designs Wen Yonggang
Ma Yuewen Modeling and Processing of Physically Based Dynamic T-splines Zheng Jianmin
Mahnaz Arvaneh Advanced Brain Computer Interface Guan Cuntai
Mao Qi Structured Prediction for Feature Selection and Performance Evaluation Tsang Wai Hung, Ivor
Mojtaba Shakeri Truck Scheduling Optimization in the Logistics of Crossdocking Stephen John Turner
Mostafa Mostafa Hashim Ellabaan Multi-Modal Memetic Framework for Locating Saddle Points, with Application to Biomolecular Systems: Water Clusters and Ring-deficient Covalently-bonded Small Molecules Ong Yew Soon
Narendar Madhavan Physical Layer Algorithms and Architectures for Accurate Detection and Classification in Cognitive Radios Vinod Achutavarrier Prasad
Nguyen Duy Duong Sensing, Allocation, and Trading of Spectral Resources in Cognitive Radio Networks A S Madhukumar
Nguyen Hai Long Advanced Classification for Streaming Time Series and Data Streams Ng Wee Keong
Nguyen Minh Luan Online Test Paper Generation for a Web-based Mathematics Testing Environment Hui Siu Cheung
Nguyen Thanh Tam Online Learning for Search and Classification Chang Kuiyu
Prabhu Kaliamoorthi Model Based Tracking of Articulated Objects Ramakrishna Kakarala
Raj Kumar Gupta Image Registration: Algorithms and Applications Deepu Rajan
Rajesh Sharma Decentralized Online Social Network Primitives Anwitaman Datta
Ren Zhixiang Exploring Effective Data Representation for Saliency Detection in Image and Video Chia Liang Tien
Salli Aruliah Arul Shalom Fast and Scalable Data Mining Algorithms for Clustering Using Graphics Processors Cong Gao
Satzoda Ravi Kumar Embedded Computing Techniques for Vision-based Lane Change Decision Aid Systems Thambipillai Srikanthan
SEAH BOON SIEW Multi-resolution Functional Summarization and Alignment of Biological Network Models Sourav S Bhowmick
Song Peng Interaction Techniques for 3D Visual Exploration on Large Displays Fu Chi Wing
Stephanus Daniel Handoko Constrained-oriented Refinement-efficacious Memetic Algorithms for Efficient Optimization of Computationally-expensive Problems Kwoh Chee Keong
Suchitra Sathyanarayana Detection of Road Markings for Advanced Driver Assistance Thambipillai Srikanthan
Sun Qian Novel Interfaces for Modeling and Painting He Ying
Teng Xiao Example-based Image Relighting Cham Tat Jen
Vittal Premachandran Exploiting Shape Properties for Improved Retrieval, Discrimination and Recognition Ramakrishna Kakarala



Contents
Name of Candidate Research Topic Supervisor
Wang Dayong Search-based Face Annotation Hoi Chu Hong
Wang Di A Hybrid Intelligent System: Genetic Algorithm and Rough Set Incorporated Neural Fuzzy Inference Network Quek Hiok Chai
Wang Lin Constraint Satisfaction Approach to Product Configuration with Cost Estimation Ng Wee Keong
Wang Zhengxiang Distance Learning Between Image and Class for Object Recognition Chia Liang Tien
Wu Pengcheng Distance Metric Learning for Multi-modal Image Retrieval and Annotation Hoi Chu Hong
Xia Yang Enhancing Network Access for Highly Mobile Nodes Yeo Chai Kiat
Yang Peng Computational Approaches for Disease Gene Identification Kwoh Chee Keong
Ying Xiang Efficient and Practical Algorithms for Discrete Geodesics He Ying
Yu Han Situation-aware Trust Management in Multi-agent Systems Miao Chun Yan
Zahra Madadi Ardekani Nonlinear Techniques for Source Detection and Localization in Shallow Ocean with Non-Gaussian Noise Annamalai Benjamin Premkumar
Zha Wei Power-aware Routing in Wireless Sensor Networks Ng Wee Keong
Zhang Wenjie Optimization of Spectrum Management in Cognitive Radio Networks Yeo Chai Kiat
Zhao Liang Characterization and Prediction of B-cell Epitopes Hoi Chu Hong
Zhao Peilin Kernel Based Online Learning Hoi Chu Hong
Zhao Wenbo Efficient Sensor Data Collection with Dynamic Traffic Patterns Tang Xueyan
Zhong Jianlong Parallel Graph Processing on Graphics Processing Units He Bingsheng
Contents
Name of Candidate Research Topic Supervisor
A. V. Subramanyam Compressed Encrypted Domain Watermarking and Forensics for Media Security Sabu Emmanuel
Agustina Real-time Collaborative Technologies and Systems for 3D Digital Media Design Over the Internet Sun Chengzheng
AlVIN NG YU-JIN Cell Profiling of Drug Treated High Content Screening Image Data Sourav S Bhowmick
Amit Kumar Singh Run-time Mapping Techniques for NoC-based Heterogeneous MPSoC Platforms Thambipillai Srikanthan
Ashish James Innovative Coding Schemes for Cooperative Multihop Radio Systems A S Madhukumar
Cheng Xiangang Local Dependencies for Near-Duplicate Retrieval Chia Liang Tien
Chua Wei Khong Watson Ontology Alignment for Knowledge Representation and Integration: Applications To Biomedical Text Kim Jung-Jae
Dao Thi Phuong Quynh Robust Discrete Geodesic and Its Applications in Digital Geometry Processing He Ying
Dilip Kumar Prasad Geometric Primitive Feature Extraction - Concepts, Algorithms, and Applications Quek Hiok Chai
Fang Yuming Visual Attention Modeling and Its Applications Lin Weisi
Farzaneh Ahmadi Voice Replacement for the Severely Speech Impaired through Sub-Ultrasonic Excitation of The Vocal Tract Goh Wooi Boon
Gao Shenghua Improving Sparse Coding with Graph, Kernel, and Structure Chia Liang Tien
Gao Wenjing A High-speed Accurate System for Phase Denoising and Unwrapping Qian Kemao
Guo Tiantian Cross-Layer Design for Video Streaming Over QoS-enabled WLANS Cai Jianfei
Li Bo View Based Techniques for 3D Model Retrieval Henry Johan
Li Fang 2D/3D Biometric and Biomedical Image Matching Using Line Constraints He Ying
Li Shukai Outlier Detection Ng Wee Keong
Li Zhenhua "Wet" Views of Protein Binding Interfaces and Hot Spots He Ying
Lin Mengda Design and Realization of a Reconfigurable Low Complexity Spectrum Detector for Cognitive Radios Vinod Achutavarrier Prasad
Liu Guoqing Challenging Issues in Classification Problems: Sparsity Control, Key Instance Detection, and Imbalanced Data Wu Jianxin
Liu Qian Computational Analysis on Protein Binding and Interactions - A Water Exclusion Perspective Hoi Chu Hong
Liu Yongchao Design and Implementation of Parallel Bioinformatics Algorithms on Heterogeneous Computing Architectures Douglas Leslie Maskell
Manish Narwaria Learning Based Signal Quality Assessment for Multimedia Communications Lin Weisi
Manju Mathew Adaptive Transmission Schemes of Cognitive Radio for Dynamic Spectrum Access Using Wavelets and Multiwavelets Annamalai Benjamin
Premkumar
Mao Yuqing Epistemology-based Social Search for Exploratory Information Seeking Sun Chengzheng
Movania Muhammad
Mobeen
Volume Graphics Shaders for GPU Lin Feng
Nguyen Ngoc Nam Self-evolving Takagi-sugeno-kang Fuzzy Neural Network Quek Hiok Chai



Contents
Name of Candidate Research Topic Supervisor
Nguyen Vu Anh Neural Network Structure for Spatio-Temporal Long-Term Memory: Theory and Applications Goh Wooi Boon
Norhaslinda Binti Kamaruddin Cognitive-inspired Speaker Affective State Profiling Quek Hiok Chai
Omid Dehzangi Discriminative Learning for Speech Recognition Chng Eng Siong
Pravin Kakar Passive Approaches for Digital Image Forgery Detection Zhang Wei
Raymond Jayaraj S/O Jayabal A Cross-layer Payload-Dropping Scheme for Improving CSMA/CA Throughput in Interference-limited Co-Channel Cells Lau Chiew Tong
Reza Khosrowabadi Modeling of Emotions Based on EEG Signals Quek Hiok Chai
Seah Chun Wei Exploiting Auxiliary Data for Designing Reliable Classifier in Domain Adaptation Ong Yew Soon
Shah Atiqur Rahman Negative Space Analysis for Human Action Recognition Ramakrishna Kakarala
Sivadon Chaisiri Resource Provisioning under Uncertainty in Cloud Computing Lee Bu Sung
Tang Chaoying Uncovering Vein Patterns from Color Skin Images for Personal Identification in Forensic Investigation Kong Wai-Kin Adams
Teng Teck Hou Cognitive Information Systems for Context-Aware Decision Support Tan Ah Hwee
Tong Rong Towards High Performance Phonotactic Feature For Spoken Language Recognition Chng Eng Siong
Umair Rafique Goal Adoption, Preference Generation and Commonsense Reasoning in Autonomous Intelligent Agents Huang Shell Ying
Vun Chan Hua Number Theory Based Signal Conversion and Processing Techniques Annamalai Benjamin Premkumar
Wan Kong Wah Search Disambiguation Techniques in Multimedia Collectiions Tan Ah Hwee
Wang Danqi Efficient User Interactivity Support for Peer-to-peer Video-on Demand Systems Yeo Chai Kiat
Wang Haixia Closed Fringe Pattern Processing Techniques: Development and Comparison Qian Kemao
Wang Kai Sketch-Based 3D Modeling and Reconstruction Zheng Jianmin
Wang Lei Audio Pattern Discovery and Retrieval Chng Eng Siong
Xiang Liu Supporting Data Collection in Wireless Sensor Networks: From Energy Conserving to Perpetual Living Luo Jun
Yang Shengbo Cooperation-aided Solutions for Data Delivery in Challenged Wireless Networks Lee Bu Sung
Zhang Da Enabling Efficient and Robust Network Service for Vehicular Users Yeo Chai Kiat
Zhang Lu Client Assignment Problems for Distributed Interactive Computing Tang Xueyan
Zhang Qing Study of Spring-Relaxation Technique for Cooperative Localization in Wireless Sensor Networks Foh Chuan Heng
Zhang Zheng Model-based Markerless Human Motion Capture from Multiple Camera Video Sequences Seah Hock Soon
Zhao Yizhi Adaptive MAC Protocols for Wireless Sensor Networks Miao Chun Yan
Zhou Hailing Towards Object-based Image Editing Zheng Jianmin
Zhu Kun Dynamic Games and Applications in Wireless Communication Networks Dusit Niyato



Contents
School of Computer Science and Engineering
A School of the College of Engineering

Nanyang Technological University
N4 Nanyang Avenue, Singapore 639798 | Tel: (65) 6790 5786 | Fax: (65) 6792 6559