The proposed project aims to develop a natural framework to solve complex optimization problems based on the principle of divide-and-conquer.

Note: Open to candidates from Computer Science or Mathematics background, full-time only.

dcku@mmu.edu.my

Photographic composition is one of the keys to the stunning shots captured by professional photographers. However, amateurs photographers are often unaware of composition rules that produce aesthetically pleasing photos and a view recommender would therefore be desirable.

Notably, automatic view recommendation is a challenging task for machine learning methods as there are infinite possible views for a photographic shot. The core objective of this research project is to build a machine learning model that can provide view recommendations or adjustments to improve the composition and aesthetics of an image. The learnt model have various useful applications including including in-camera guide, robot photography and image cropping or thumbnailing.

lkwong@mmu.edu.my

The project is related to positioning application development that will support user to extract real-time location information. The project requires to use optimization solution to improve the accuracy of the location information.

kgtan@mmu.edu.my

The work required modelling of the characteristics of indoor signal in time and spatial domain. The scope of the study include measurement and theoretical works.

kgtan@mmu.edu.my

The work covers modeling, prototyping, and measurement.

kgtan@mmu.edu.my

Big Data Analytics and Deep Learning are two high-focus of data science. Big Data has become important as many organizations both public and private have been collecting massive amounts of domain-specific information, which can contain useful information about problems such as national intelligence, cybersecurity, fraud detection, marketing, and medical informatics. Companies such as Google and Microsoft are analyzing large volumes of data for business analysis and decisions, impacting existing and future technology. Deep Learning algorithms extract high-level, complex abstractions as data representations through a hierarchical learning process. Complex abstractions are learned in a given level based on relatively simpler abstractions formulated in the preceding level in the hierarchy. A key benefit of Deep Learning is the analysis and learning of massive amounts of unsupervised data, making it a valuable tool for Big Data Analytics where raw data is largely unlabeled and uncategorized. This project aims to explore how Deep Learning can be utilized for addressing some important problems in Big Data Analytics, including extracting complex patterns from massive volumes of data, semantic indexing, data tagging, fast information retrieval, and simplifying discriminative tasks. This project also aims to investigate some aspects of Deep Learning research that need further exploration to incorporate specific challenges introduced by Big Data Analytics, including streaming data, high-dimensional data, scalability of models, and distributed computing.

shohel.sayeed@mmu.edu.my

Living in the era of big data, it has been witnessing the dramatic growth of heterogeneous data, which consists of a complex set of cross-media content, such as text, images, videos, audio, graphics, time series sequences, and so on. Such hybrid data comes from multiple sources and hence embodies different feature spaces. This situation is creating new challenges for the design of effective algorithms and developing generalized frameworks to meet heterogeneous computing requirements. Meanwhile, deep learning is revolutionizing diverse key application areas, such as speech recognition, object detection, image classification, and machine translation, with its data-driven representation learning. Thus, it has become critical to explore advanced deep learning techniques for heterogeneous big data analytics, including data acquisition, feature representation, time series analysis, knowledge understanding, and semantic modeling. This project will mainly focus on deep learning and cross-media methods for Big Data representation such as data-driven feature learning via deep learning methods, large-scale multimodal data acquisition, novel datasets and benchmarks for heterogeneous big data analytics, multimodal information fusion via deep learning and so on.

shohel.sayeed@mmu.edu.my

This research integrates research findings on cognitive, motivational and social-emotional aspects of student engagement to formulate instructional strategies for incorporating multidimensional characteristic analysis of learners into personalized learning. To verify the effectiveness of the proposed study framework, profiles of actual learners will be constructed, with a higher learning institution students as subjects. The participants will be divided into 2 groups (control and experimental) to compare the effectiveness of the learning experiences. The outcome from this research is a multidimensional characteristic learner profiling framework that puts the learner at the center, providing assessment and instruction that are tailored to students’ particular learning and motivational needs. The proposed framework will contribute to the society by humanizing the innovative online learning through Malaysian Higher Education 4.0.

Research Cluster: Education & Knowledgeable Civil Society

lau.siong.hoe@mmu.edu.my

Big Data Analytics and Deep Learning are two high-focus of data science. Big Data has become important as many organizations both public and private have been collecting massive amounts of domain-specific information, which can contain useful information about problems such as national intelligence, cybersecurity, fraud detection, marketing, and medical informatics. Companies such as Google and Microsoft are analyzing large volumes of data for business analysis and decisions, impacting existing and future technology. Deep Learning algorithms extract high-level, complex abstractions as data representations through a hierarchical learning process. Complex abstractions are learned in a given level based on relatively simpler abstractions formulated in the preceding level in the hierarchy. A key benefit of Deep Learning is the analysis and learning of massive amounts of unsupervised data, making it a valuable tool for Big Data Analytics where raw data is largely unlabeled and uncategorized. This project aims to explore how Deep Learning can be utilized for addressing some important problems in Big Data Analytics, including extracting complex patterns from massive volumes of data, semantic indexing, data tagging, fast information retrieval, and simplifying discriminative tasks. This project also aims to investigate some aspects of Deep Learning research that need further exploration to incorporate specific challenges introduced by Big Data Analytics, including streaming data, high-dimensional data, scalability of models, and distributed computing.

shohel.sayeed@mmu.edu.my

Living in the era of big data, it has been witnessing the dramatic growth of heterogeneous data, which consists of a complex set of cross-media content, such as text, images, videos, audio, graphics, time series sequences, and so on. Such hybrid data comes from multiple sources and hence embodies different feature spaces. This situation is creating new challenges for the design of effective algorithms and developing generalized frameworks to meet heterogeneous computing requirements. Meanwhile, deep learning is revolutionizing diverse key application areas, such as speech recognition, object detection, image classification, and machine translation, with its data-driven representation learning. Thus, it has become critical to explore advanced deep learning techniques for heterogeneous big data analytics, including data acquisition, feature representation, time series analysis, knowledge understanding, and semantic modeling. This project will mainly focus on deep learning and cross-media methods for Big Data representation such as data-driven feature learning via deep learning methods, large-scale multimodal data acquisition, novel datasets and benchmarks for heterogeneous big data analytics, multimodal information fusion via deep learning and so on.

shohel.sayeed@mmu.edu.my

High volume of data available due to sensors’ ubiquity and pervasiveness need to be processed and analyzed to extract meaningful or actionable information to recommend appropriate changes in IoT systems. Due to the sheer volume of the data from these IoT systems, any errors from user entry, data corruption, data accumulation, data integration, or data processing can be snowball, causing massive errors that can detrimentally affect the decision-making process. It is found from the literature review that incomplete (missing values), duplicate, inconsistency and inaccurate are the four common dimensions of dirty data exists. Consequently, there needs to be a clear understanding of the challenges associated with data quality and a way to evaluate and ensure that data quality in Iot Data Analytics (DA) is maintained for different applications. Although several studies have made various proposals on managing data quality (DQ), there seems to be a lack of methodologies that are general enough to assist developers in managing the quality of the data for the IoT systems. Therefore, there is a need for a complete data quality framework that will clean intelligently. Machine Learning (ML) technique will be used to overcome the issues.

jakir.hossen@mmu.edu.my

The recent proliferation of the Internet of Things (IoT) has enabled seamless integration of interconnected sensors, actuators, and any other computational devices, in the form of distributed computing network. The establishment of such network requires a form of identification mechanism for all the devices to be connected. Potential threats exist, such as rogue device that can masquerade authentic IoT device, enabling it to take control over the entire network. Existing implementation relies on a trusted third-party authentication and identification management that resides over the cloud or within the distributed network. The deployment of such auspicious solution has faced many challenges since the centralization of the trust and connectivity of the IoT devices forces network to become a single point of failure that may disrupt the entire IoT operations. A decentralized approach offers an elegant solution to solve this problem. Hence, this work proposes a permissioned blockchain-based identity management framework for interconnected IoT devices within a distributed network infrastructure. Unlike existing approach where the identification and authentication of devices are handled by a single entity, this formulated framework utilizes the immutable blockchain network as a decentralized identity management authority and repository through a combination of distributed ledger scheme and smart contracts.

nazrul.muhaimin@mmu.edu.my

This project proposes a permissioned blockchain-based framework for the IoT devices to elastically offload its critical and intensive computation task to nearby cloud and to communicate securely between them without authorization, authentication and identification by the cloud. Unlike existing IoT-cloud framework where IoT to IoT communication is practically impossible, this framework augments the possibilities of cooperation, sharing and messaging between the IoT devices that unleash the realization of totally new applications.

nazrul.muhaimin@mmu.edu.my

This project proposes to develop a privacy preserving decentralised architecture using deep learning and blockchain technology. The proposed architecture will provide data confidentiality and other important security properties to ensure confidence and wider adoption of distributed deep learning in disciplines such as healthcare where data privacy is of utmost importance.

shheng@mmu.edu.my

In order to provide secure message communication within mobile networks, encryption scheme with keyword search schemes that enable searching of keywords within encrypted message are desirable. These schemes protect the confidentiality of encrypted messages but at the same time allow intermediary gateways to search encrypted messages for any keywords as instructed by the receiving mobile nodes. The main objective of this research is to design and analyse a secure and efficient searchable encryption scheme with additional features which could be operated using different platforms.

shheng@mmu.edu.my

Despite the popularity of mobile cloud and cyber-physical cloud systems, security issues in untrusted cloud environment and physical devices remain major concerns. Data sharing protocol is a potential cryptographic solution that can be used to facilitate secure data sharing in a cyber-physical cloud environment. The protocol is designed to achieve authentication between a physical device and the cloud controller, and provide a secure end-to-end secure communication in the cloud. This research focuses on the design and analysis of secure data sharing protocol via mathematical approach.

shheng@mmu.edu.my

One of the most requirements in Industry 4.0 is predictive maintenance (PdM). PdM methods helps to assist early detection before any failures and errors in machinery reaches to critical stage. Anomaly detection methods are able to focus on the core requirement PdM to find anomalies in the working equipment at early stages and alerting the manufacturing supervisor to carry out maintenance activity. However, there is high false-positive rate is a long-standing challenge for anomaly detection algorithms, especially in high-stake applications. To identify the true anomalies, in practice, analysts or domain experts will be employed to investigate the top instances one by one in a ranked list of anomalies identified by an anomaly detection system. The main objective of the proposal is to help the analyst to discover more true anomalies given a time budget. An active anomaly detection approach is proposed for PdM. To automatically tune the anomaly detectors to maximize the number of true anomalies discovered. several algorithms for active learning with tree-based AD ensembles and reinforcement learning techniques will be used to explicitly optimize the number of discovered anomalies. This might reduce the time and costs associated to labeling while delivering the same or similar anomaly detection performances. The techniques will be implemented in a way that it will be adaptable for different fields and address different industrial use cases, helping stakeholders gain trust and thus promote the adoption of data-driven PdM systems in smart factories.

jakir.hossen@mmu.edu.my