Research projects in Information Technology

Current federated learning architectures in mobile healthcare are limited to a centralised model without considering the full continuum of mobile-edge-cloud. Additionally, to support different data privacy needs of patients as well as the limitations of mobile environments, there is a need for considering a multi-level federated learning architecture for the mobile-edge-cloud continuum.

Chronic conditions are becoming a serious global and national health problem. Recommendation systems play an important role in supporting patients in managing their long-term health issues. They generally rely on expert rules or machine learning models to provide health advice. Recently, generative AI tools, such as ChatGPT, have become a popular focus of research. In healthcare, they show strong potential to facilitate the process of generating health-related advice without the need for predefined rules or training data. Yet, their reliability remains a serious concern. 

Diabetes can be effectively controlled by maintaining a healthy diet, well-managed blood glucose level and regular physical activity. Evidence suggests that improving dietary habits can play a crucial role in preventing the onset or progression of diabetes. A large number of mobile apps have been recently introduced to assist individuals with self-management of diabetes.

The future of autonomous driving systems holds great promise, offering a solution to address the challenges associated with human errors and the mental fatigue of driving. However, there are trade-offs between the modularity (henceforth interpretability) and the efficiency in existing end-to-end modular autonomous driving models. In this PhD project, student is expected to conduct research in the area of end-to-end modular autonomous driving using computer vison and deep learning methods.

This project will explore the security and privacy challenges inherent in the tokenization of real-world assets (RWAs) using the blockchain technology. As industries increasingly adopt tokenization to digitize and trade assets like real estate, commodities, and fine art, ensuring the security and privacy of these transactions becomes critical. The research will focus on identifying vulnerabilities in existing tokenization frameworks, analyzing potential risks, and developing novel security protocols to protect sensitive data and ensure the integrity of tokenized assets.

This project focuses on developing algorithms capable of automatically identifying and categorizing mobile ringtones. This involves leveraging machine learning techniques to analyze audio signals from mobile devices and classify them into different categories or types of ringtones. The activities of the project include gathering a diverse dataset of audio samples representing various types of mobile ringtones.

In recent years, the rise in cybercrimes has significantly increased the vulnerability of the open internet to various threats and cyber-attacks. Among these, phishing stands out as one of the most perilous crimes worldwide. In a phishing attack, perpetrators create fraudulent websites that mimic legitimate ones (such as fake bank websites). These deceptive sites lure users into disclosing sensitive financial, personal, and confidential information.

The detection of human activities is crucial for effective monitoring purposes. The challenge lies in accurately and promptly identifying various types of activities from videos and images captured in diverse, real-world environments. Both classical machine learning methods and deep learning techniques can be employed to tackle this task.

Deepfakes, derived from "deep learning" and "fake," involve techniques that merge the face images of a target person with a video of a different source person. This process creates videos where the target person appears to be performing actions or speaking as the source person. In a broader context, deepfakes encompass other categories such as lip-sync and puppet-master. Lip-sync deepfakes alter videos to synchronize mouth movements with a provided audio track.

This project aims to develop robust algorithms capable of identifying and analyzing fingerprints extracted from both static images and video footage. Machine learning techniques, particularly computer vision and pattern recognition methods, will be utilized to automate the process of fingerprint detection. These methods will be trained to learn patterns from fingerprint features and detect them using object detection approaches. A dataset of fingerprint images and videos, annotated with ground truth information will be collected.