Human-Computer Interaction Lab

Recent fires in the Sichuan Forests and a Shanghai factory resulted in over 30 firefighter deaths due to unpredicted sudden change of environment. In fact, the number of firefighter fatalities was significantly higher in China despite fire rate in USA is being 80% higher than in China. It is reported that US firefighters have access to better protective equipment and training than their Chinese counterparts, which may explain the differences in fatality rate. In today frontline firefighting practices, Cyber-Physical Systems (CPS) is formed by utilizing a variety of sensors and big data analytics to develop firefighter equipment and personal protective equipment (PPE) for improving safety and effectiveness of firefighting. These technologies facilitate the development of Smart FireFighting (SFF) when integrated with sensors that coupled with wireless communication technologies.

The aim of these projects is to design a smart firefighting operations system (SFOS) with the vision for enhancing firefighting activities. The SFOS vision includes the collection and integration of information from wireless sensor technology, computational tools to analyze that information to perform safety assessment of firefighter and building models for indoor localization and mapping. The realization of this vision would improve the firefighter occupational health and safety, allow firefighter to coordinate better with each other and firefighters to execute fireground operations more effectively. The key research topics for Ph.D. study including, but not limited to:

1. Design of Smart Wearable Embedded Sensors in Firefighter Personal Protective Equipment.
2. A Novel 5G/6G Wireless Communication Framework for Smart IoT Firefighting.
3. Design of Wireless Sensor Network for Firefighter Remote Safety Monitoring.
4. Design of AR Headset for Information Visualization for Firefighters.
5. Indoor Thermal Imaging Analysis for Obstacles and Dangerous Items Detection.
6. Design of Smart Communication Device for Firefighters Data Transmission.
7. Firefighter Safety Risk Assessment with Sensors Fusion.
8. Optimization and Synchronization of Multi-Drones in Fire Scene Remote Investigation.
9. Indoor Localization and Mapping in Firefighting Under Smoke-Filled Environment.
10. Study of Human Factors in Firefighting with Sensors.
11. Study of Human Factors in Firefighting with MR and AR.
12. Identification and Simulation of Fire Traces with Computer Vision.

The Ph.D. candidates should have basic research experience in some of the following relevant areas of study, including but not limited to computer vision, sensor hardware design (PCB design), wireless communication technology, AR / VR / MR technology (Viro, Unity3D, ARCore), health analytics, indoor localization and mapping and other related fields. Prefer with Computer Science, Electronic and Electrical Engineering, Robotics and relevant backgrounds.

Financial fraud remains a challenging problem for financial institutions in China as well as more widely internationally. Types of fraud include email phishing, payment fraud, theft identity, fake account information, and document forgery. Popular existing fraud detection models use machine learning including supervised learning algorithms (logistic regression, decision trees, random forests, and neural network), and unsupervised learning algorithms (k-means clustering and local outlier factor). These algorithms are used to detect the anomalies and abnormality in customers’ behavioural patterns.

Nonetheless, the effectiveness of a fraud detection system requires constant updates and improvements of a ML-powered model. Fraudsters always come up with new tricks to game the system and reliable ML-powered model should be able to tackle such problems. Moreover, ML-powered models are data-hungry and lots of data are essential to train the model in predicting diverse and new fraud activities. The difficulty increases when the data are challenging to obtain. Therefore, this project is intended to propose and explore two solutions to address these specific issues.

• Firstly, apply semi-supervised learning to artificially generate data to imitate the under-represented fraud data (oversampling) to overcome the imbalanced and shortage of data.
• Secondly, is to explore the use of reinforcement learning models to develop fraud detection strategies and self-improve fraud detection over time, and automatically detect abnormal customers’ behaviour within specific transactional contexts.
• Thirdly, considering that most of the user operations happen on mobile devices, it worthies investigating the prevention of the fraudulent behaviors at the edge devices, by exploring efficient yet intelligent solutions on user intent prediction.

The work will be conducted over 4 years with publicly available individual transaction and corporate data, as well as privately donated data that will be acquired during the project. The research student will have excellent analytic skills, as well as an interest in business applications.