Proposed Project Title: Online performance and energy optimisation

Principal Supervisor: Prof Hans Vandierendonck

Project Description: Dynamic Voltage and Frequency Scaling (DVFS) uses processor controls to change operating voltage and frequency. These parameters directly determine both power consumption and performance. Overall energy savings can be achieved by judiciously adjusting processor frequency as programs move through different phases of compute-intensive and memory-bound computations. What type of a phase a program is in can be determined by monitoring Hardware Performance Monitoring (HPM) counters, which are today present in all common processor architectures. The state of the art approach in DVFS builds machine learning models that connect HPM measurements with desired DVFS settings. This requires substantial off-line analysis of the program, which in itself may consume significant energy. There is also a risk that the model is specific to the data set that is processed, where changes in the data would cause the model to select inappropriate DVFS settings.

The goal of this project is to evaluate controlling DVFS using online machine learning. Rather than constructing a model off-line, the DVFS controller can try different settings during execution and take a number of samples during program execution. These samples are used to train (or fine-tune) the online model. This project requires skills in machine learning and operating systems and computing systems aspects. A plausible implementation could leverage the power governor interface of the Linux operating system.

Literature:

[1] Linux frequency scaling: https://wiki.archlinux.org/title/CPU_frequency_scaling

[2] A portable library for collecting HPM: https://variorum.readthedocs.io/

[3] A survey on software methods for energy reduction, including DVFS: https://doi.org/10.1177/1094342016665471

[4] A proposition for joint DVFS and concurrency control: https://dl.acm.org/doi/10.1145/1454115.1454151

Objectives:

• Setting up a number of workloads to experimentally evaluate DVFS
• Characterising the workloads in terms of the potential of DVFS
• Selecting and applying relevant online machine learning models for DVFS control
• Implementing an operating system power governor to implement the DVFS policy

 

Proposed Project Title: Running Microsoft SEAL on Raspberry Pi for Secure IoT-based Data Analytics

Principal Supervisor: Dr Amir Sabbagh Molahosseini

Project Description: Project Description:

Homomorphic encryption is an encryption method that enables performing computations on encrypted data without the need to decrypt the data. Therefore, data can be encrypted by the user or low-end Internet-of-things (IoT) device and then will be sent to a third party for performing computations on the encrypted data without the content of the operands. Then, the computation result will be sent back to the sender or any other party with access to the key for decryption. This process allows fully privacy-preserving and secure data analytics. The Microsoft SEAL is an open-source and free library that allows to perform computation on encrypted data.

The target of this project is to install and run Microsoft SEAL on the Raspberry Pi board. In other words, this project aims to create an IoT node based on Raspberry Pi to sense data, and then encrypt the data, and send the encrypted data to a server/cloud for computations. Therefore, the computations on encrypted data will be performed on the server, and the computation results that are in the encrypted format will be sent back to the IoT node for decryption that needs to be done by Raspberry Pi. The performance in terms of running time and memory usage needs to be analyzed to detect the bottlenecks improve the implementation, and try to find ways to reduce IoT/cloud communication data.

The available Raspberry Pi boards in the Innovation-by-Design-Lab will be used for designing the required IoT node to encrypt/decrypt the data. Besides, the EEECS HPDC system can also be used for initial experimentations and code development. A knowledge of C/C++ programming is necessary for this project.

References:

3. N. Smart, "Computing on Encrypted Data," IEEE Security & Privacy, vol. 21, no. 4, pp. 94-98, 2023. (https://ieeexplore.ieee.org/document/10194492)
4. N. Downlin, et. Al., “Manual for Using Homomorphic Encryption for Bioinformatics”, Microsoft Research, 2020. (https://www.microsoft.com/en-us/research/wp-content/uploads/2015/11/ManualHE-3.pdf)
5. Microsoft SEAL Homomorphic Encryption Library, (https://github.com/microsoft/SEAL)

Objectives: The goal of this project is to implement the Microsoft SEAL software library on Raspberry Pi boards, and the specific objectives are:

• Study relevant literature on homomorphic encryption
• Study some basics of HE operations and work with an existing library in C++, Microsoft SEAL.
• Experimentally evaluate the performance of Microsoft SEAL on Raspberry Pi boards

 

Proposed Project Title: Security in Versal AI core which is the most powerful reconfigurable chip till date.

Principal Supervisor:  Dr Arnab Kumar Biswas

Project Description: If you already know any hardware description language, your task will be to design and test secure digital design using existing Network-on-Chip in our ADK-VA600 hardware platform. This has an AMD-Xilinx Versal AI core which allows us to use the adaptable SoC platform more efficiently and implement something new.

If you don’t know any hardware development language, you will work on high level synthesis using the Xilinx provided tool and continue design and development of secure digital design.

The work aim is to explore different types of attacks possible in the platform using existing applications or benchmark programs. It will be great if we can identify new attacks and countermeasures for those attacks.

Objectives: 

1. First objective is to learn about the Versal AI core and how to use the ADK-VA600 hardware platform efficiently
2. Second objective is to identify existing attacks that can be launched in the system
3. Implement the system using existing hardware/software solutions to launch and evaluate the attacks
4. Find out new attacks (optional)
5. Find countermeasures for the attacks
   
   

Proposed Project Title: Reconfigurable Intelligent Surface for improving Wireless Communication coverage and remote Sensing

Principal Supervisor: Dr Dmitry Zelenchuk

Project Description: There is a great demand for low-cost microwave surfaces that can be concealed within normal environments and help enhance wireless communications. One of the challenges is to design low-cost reflecting and/or redirecting surfaces in the vicinity of mobile base station antennas to prevent interference or direct energy into a shadow area. These are used in applications spanning from current WiFi networks to 5G and beyond.

Currently, such a reconfigurable intelligent surface has been developed in QUB and is undergoing experimental trials in ECIT. The student will have to learn how to program the reconfigurable intelligent surface and measure its performance. By measuring various indoor and outdoor scenarios the student will develop strategies for maximizing the wireless coverage as well as explore the ways to use the devices for remote sensing.

Objectives: 

Program and measure mm-wave reconfigurable intelligent surfaces in various indoor and outdoor environments to achieve better coverage and explore remote sensing using wireless channels.

 

Proposed Project Title: The Code Breaker: Attacking the security of systems via Fault Injection Attacks

Principal Supervisor: Dr. Ayesha Khalid

Project Description: FPGA (Field-Programmable Gate Array) based cryptographic implementations are used in a wide range of applications across various industries due to their flexibility, high processing power, and suitability for accelerating cryptographic algorithms. However, as these FPGA configurations are stored as files called bitstreams, they become potential targets for malicious adversaries seeking to exploit vulnerabilities in the implementation. Fault Injection Attacks (BiFI) are a class of attacks that aim to manipulate or inject faults into the configuration bitstream of an FPGA [1]. The primary goal of BiFI attacks is to compromise the security or functionality of an FPGA-based system by altering the configuration bitstream that defines the hardware logic of the FPGA. The figure below gives an overview of how BiFI attack works:

This internship opportunity will provide the student a hands-on experience to learn the use of FPGA boards. The student will be provided with an AMD Xilinx Zynq Z7020 FPGA board on a PYNQ-Z2 board (see picture) [2]. The student will keep the board during the internship, the toolchain for these simple boards can be installed on your laptop and they come with a single USB connection for both programming and debugging. This internship would provide a valuable learning experience for the student, as they would learn to use the toolchain to program the board and implement simple cryptography algorithms. This internship explores the Bitstream Fault Injection Attacks (BiFI) on FPGA bitstreams, specifically focusing on cryptographic implementations of a block cipher called the AES. AES stands for Advanced Encryption Standard. It is a widely used symmetric-key encryption algorithm that is employed to secure sensitive data. The project studies the common attack strategies and vulnerabilities as an attacker seeks to manipulate or inject faults into the FPGA bitstream of any block cipher.

[1] Swierczynski, Pawel, et al. "Bitstream fault injections (BiFI)–automated fault attacks against SRAM-based FPGAs." IEEE Transactions on Computers 67.3 (2017): 348-360.

[2] http://www.pynq.io/board.html

 

 

Objectives: This project will give you confidence in working with new technologies and tools, enhancing your problem-solving skills. You will learn traits that can lead to a potential career path in future, i.e., effective documentation and analysis of your findings related to the project, develop communication and presentation skills, learn how to work effectively within a professional team, etc. Some technical objectives are listed below.

1. To gain practical experience of working with AMD Xilinx Zynq Z7020 FPGA board on a PYNQ-Z2 boards. You will learn the use of the toolchain to program simple algorithms on board.
   
   
2. Learn and appreciate the importance of cybersecurity in modern engineering projects and learn basic cryptography algorithms along the way. Understand the operation of block ciphers and their implementations on FPGAs (several open-source Verilog HDL implementations are available).
   
   
3. To create a practical experimental setup on an FPGA board for injecting faults into bitstreams containing AES implementations. Launch the reported fault injection attacks on FPGA bitstreams and evaluate their impact on the AES security compromise.
   
   
4. Learn to benchmark an algorithm to evaluate its suitability on FPGA devices. Exploring optimizations that may positively affect the performance or cost of an algorithm.
   
   
5. Gain insights into security challenges the FPGA devices face and the appropriate countermeasures and strategies to protect them. Contribute to the broader field of cybersecurity, FPGA security and cryptanalysis, making valuable contributions to the ongoing efforts to secure digital systems.

 

Proposed Project Title: Stable phase reference signal for ultra-large phased arrays for the solar space satellite

Principal Supervisor - Dr Neil Buchanan

Project Description: The solar power space satellite (SPSS) concept aims to harness solar energy from space via a geostationary satellite and transmit it to Earth using microwave beams. It has the potential to generate tens of gigawatts of uninterrupted clean energy, reducing reliance on fossil fuels. A recent UK Government feasibility study [1] confirmed the viability of the SPSS with modern technology, projecting a satellite launch by 2040 with the necessary investment. The Centre for Wireless Innovation (CWI) at QUB is already working on the SPSS with a number of high profile industrial partners.

The huge solar power space satellite (up to 2km across) will rely on a “retrodirective” antenna to accurately point the microwave beam to earth. Pointing accuracy of hundredths of a degree are needed. To achieve this, each of the billions of antennas on the satellite require a stable phase reference which must be efficiently distributed around the entire satellite structure. The focus of the internship is a practical demonstration of a scale model of the stable phase reference. This involves the use of demo PCBs, showing that the concept works practically. This is an important demonstration that we will show to potential stakeholders to secure further research funding.

[1] https://www.gov.uk/government/publications/space-based-solar-power-de-risking-the-pathway-to-net-zero

Objectives:

• Put together a practical demonstration of a stable phase reference system for at least four antennas. The concept has already been proven by simulation.
• The building block of the stable phase reference will be an IC from analog devices, which is available as an evaluation board, allowing the experimentation to start fairly quickly at the start of the internship. Phase stability will be determined using a 4 channel oscilloscope.
• Some knowledge of practical electronics is required, knowledge of radio/microwave particularly welcomed.

The internship will provide an excellent opportunity for engagement with relevant industry via several live research projects.

Proposed Project Title: Integrating PowerBI with 3D Unity Dashboards for Enhanced Corporate Data Visualisation 

Principal Supervisor: Prof Karen Rafferty

Project Description: This summer internship offers an exciting opportunity to work within the Advanced Research and Engineering Centre (ARC) at EEECS, in collaboration with ARC teams at PwC and Ulster University. The successful intern will join our Virtual Reality for Corporate Systems project team to investigate the feasibility of integrating data from PowerBI to 3D Unity-generated environments.

The primary focus of the internship project will be on assisting in the creation of a proof-of-concept interactive dashboard that allows users to visualise and interact with data in a virtual space. By leveraging Unity’s 3D capabilities, this project explores how traditional, static data presentations might be transformed into engaging and dynamic visualisations within the corporate environment.

Objectives: 

• Gain an understanding of the processes involved in linking PowerBI data with Unity and apply these techniques to import data into a 3D Unity environment.
• Assist in the creation of effective Unity-based dashboards to support the wider project’s objectives on enhancing data display, user understanding and engagement through interactive visualisations.
• Overcome technical challenges and refine the dashboard design through effective collaboration and regular meetings both within the ARC research team at EEECS and with technical leads at PwC

 

Proposed Project Title: Exploring security vulnerabilities in porting Unmanned Aerial System libraries over CHERI Arm Morello Board

Principal Supervisor: Dr Vishal Sharma

Project Description Unmanned Aerial System relies on ground terminals for the successful execution of missions and transfer executables into the drone's firmware. The memory management capabilities of RISC architecture may be advantageous for such requirements. Using these capabilities, for instance, we intend to separate the code based on the security vulnerability of the various regions of the firmware and determine if executing it over the terminal controlling the UAVs is secure against known attacks.

The project intends to exploit and report on the capabilities of CHERI architecture as they pertain to the impact of utilising drone libraries, commanding drones, and any significant vulnerabilities that can expose the ground terminals. In addition, we will conduct a feasibility study on introducing drones and security libraries to CHERI architecture, as well as the associated overheads using a physical ARM Morello board.

Objectives:

1. UAS Code porting over CheriBSD on Arm Morello board.
2. Explore porting errors and identify potential vulnerabilities resultant because of the porting.
3. Reporting of the project findings and recommendations on execution cost of the existing UAS libraries.

 

Proposed Project Title: SAM-otater: Using the latest AI to automatically annotate images in medicine and biological sciences

Principal Supervisor: Dr Richard Gault

Project Description: Segmentation is the process of automatically isolating regions of an image or video for further inspection, for example, identifying all cells in a microscopy image or tracking the movement of an animal in a video. When there are lots of objects to identify and track, or the objects have a complex shape, it can be very time-consuming for scientists to manually annotate images and videos. This can also lead to human error in the annotations. In 2023, Meta released “Segment Anything Model” or SAM for short which significantly changed the landscape of segmentation in image analysis. This has help to unlock new avenues of research for scientists to increase our understanding of biological systems by reducing the amount of time required to manually annotate/segment images.

This project will extend an existing QUB project to incorporate the SAM model as an annotation aid within a Python based GUI. This project will provide the opportunity to develop skills in Python, AI, Deep Learning, Computer Vision, and collaborate with biologists and scientists from experimental medicine who will act as “end -users” of the computational tool.

Objectives:

Incorporate the SAM model in to an existing Python toolbox so that users can automatically annotate images and videos through an intuitive Graphical User Interface.

 

Proposed Project Title: Classification of Natural Questions based on their Information Needs

Principal Supervisor: Prof Karen Rafferty and co-supervisor Dr Muhammad Roman

Project Description: Natural language questions asked on a document can be answered from a single, or multiple passages depending on the nature of the question. Simple questions have a single relation r with an entity e, r(e), for example, Invented(Telephone). For instance, the question "Who invented the telephone?" falls under this category. Intersection queries look for the knowledge coming from two intersecting simple queries, r₁(e₁) Ո r₂(e₂). For example, “Which project had Chris as PI and Barry as CI”. Composition questions on the other hand require multiple passages to be retrieved in order to answer a single question. A composition question has a relation r, such that r₁(r_i(e_i)). It is important to understand the type of question in Open Domain Questions Answering (ODQA) to get the right information extracted from relevant documents. Different ODQA outperforms for a particular type of question but is not good for other types of questions as one solution does not fit all. Natural language understanding (NLU) plays a vital role in understanding the question's context and type. By understanding the underlying question template we can use some deep learning techniques to categorise the question into one of the simple, intersection, or composition classes. A number of ODQA datasets including NQ and QAMPARI can be used to train and evaluate the classifiers. The classification step can be used as a preliminary step for selecting among Passage Independent Generator (PIG) and Fusion-in-Decoder(FiD) methods, including Retrieval Augmented Generation (RAG) for generating update and private knowledge from documents.

The project can give a good understanding of natural language processing and natural language understanding to new researchers. It will also give a hands-on experience with deep learning models, especially for text classification which can be broadly applied to other tasks including sentiment analysis.

Objectives:

• To understand the relation among entities in natural language questions
• To classify natural language questions as simple, intersection, or composition based on the relation and the required passages to answer such questions
• To investigate the capabilities of various deep learning approaches in capturing the linguistic structure from natural questions

To improve existing models and enhance their capabilities of classifying natural language questions

 

Proposed Project Title: Use of Mixed Reality for controlling a steel structure assembly robot

Principal Supervisor: Prof Karen Rafferty and co-supervisor Dr Miftahur Rahman

Project Description: : ‘Assembly and robotics Innovation in Steel Building Erection’ (ARISE) is a research project to develop a process for robotically assembling structural steel connections. This is US-Ireland multinational collaborative project between New York University, University of Texas San Antonio, University of Galway and Queen’s University Belfast. In Queen’s, both EEECS and Civil Engineering are working collaboratively to enable robotic assembly of a structural steel frame.  This is being achieved by creating a computer vision model using simulation scenarios and machine learning and then using Mixed Reality to control a manipulator for small scale trials. The simulation scenario was created using Unity Game Engine and ROS is used as a middleware. Mixed reality will enable the safe and interactive usage of robotic manipulator later full scale robotic structural steel assembly.

Objectives:

1. Review the state-of-the-art mixed reality systems
2. Find and investigate the usage of mixed reality for robot control in simulation and real-world
3. Setup mixed reality systems to control a robot inside a simulation scene in Unity with ROS

 

Proposed Project Title: VR Lab Research Intern

Principal Supervisor: Dr Matthew Collins

Project Description: Opportunity to work on a variety of projects with a range of VR and AR hardware. Gaining valuable insight into postgraduate education through exposure to MSc software development projects.

Primary goals will be building on the work of previous VR lab intern who has been working on VR and AR apps particularly in Mixed Reality areas producing prototype experiences for use by colleagues in Bio Sciences/Social Sciences, Education & Social Work among others.

This falls under multi-multidisciplinary research on both a computing, software development, education domain and a number of other disciplines.

Objectives: 

• requirements analysis and exploration of problem domain(s) – e.g. examining an existing VR protype for use in OCD exposure therapy, proposing potential areas where newer technologies could enhance and provide additional opportunities
  
  
• development of a prototype Mixed Reality app suitable for use in research by colleagues in the Centre for Technological Innovation in Mental Health and Education (TIME Centre) at QUB.
  
  
• app would likely be developed for Meta Quest but exact platform is open (potential for more ambitious directions to be explored – Multi device support etc.)
  
  
• a successful prototype developed at this stage could perhaps lay the ground work for a more ambitious final year project in the future

 

Proposed Project Title: Fast & Furious with self-drive AI

Principal Supervisor: - Dr Yun Wu

Project Description: Simultaneous Localization and Mapping (SLAM) is one of the key enabling technology of autonomy for robotics and unmanned vehicles. However, it require huge computational workload to process the sensor perceptions and make decision using complex optimization algorithms. It is a big challenge to deploy SLAM on micro robotics or vehicles, which has limited resources in computation and power supply, especially those using battery-powered devices. Approximate computing is one of the trending computing paradigm that enables saving in computing cost, power, and space. In this project, students are going to use JetRacer AI platform to deploy SLAM for autonomous vehicle applications, using the camera and LiDar sensors. By applying approximate computing to SLAM, it is going to explore and estimate how much performance the SLAM can achieve and how much resources in terms of computing cost and power it takes. It is hoped that this will build up the understanding of both SLAM principle and approximate computing and strengthen the programming skills in C/C++ and Python.

Objectives: 

1. To be familiar with JetRacer AI platform
2. To be familiar with SLAM using JetRacer
3. To learn approximate computing techniques and apply to SLAM on JetRacer