We invite applications for a PhD award linked to the AdSoLve project Addressing socio-technical limitations of Large Language Models (LLMs) in medical and social contexts. AdSoLve is a large multi-disciplinary project funded by UKRI and RAi UK (for details, see www.rai.ac.uk). AdSoLve is led by Prof Maria Liakata at QMUL and involves four University partners (QMUL, Nottingham, Sheffield and Warwick) and 21 external partners, including commercial companies, NHS trusts, NHS England and UK AI hubs and CDTs.

AdSoLve aims to: 1) create evaluation benchmarks (including novel criteria, metrics and tasks) for assessing ethics and safety risks of LLMs in real world applications; 2) draw on expertise in law, healthcare, AI ethics and safety to devise new machine learning methodologies to mitigate these risks. The project may also involve developing modules for temporal reasoning and situational awareness in long-form text, dialogue and multi-modal data, as well as alignment with human preferences, bias reduction and privacy preservation.

We are seeking a PhD candidate in machine learning theory (statistical learning theory and deep learning theory) or theoretical-oriented topics, e.g., trustworthy machine learning, efficient machine learning. The project aims to theoretically understand why ML models perform well and/or design efficient and robust algorithms in trustworthy machine learning. The topics include but not limited to:

1. Statistical-computational gap in modern machine learning

2. Robustness of neural networks for trustworthy ML systems

3. Fine-tuning of modern machine learning models

The successful candidate is expected to have a solid background in applied mathematics/statistics/computer science or related discipline. Advanced coding skills are a big plus.

For further details please contact Dr. Fanghui Liu (fanghui.liu@warwick.ac.uk), Assistant Professor In Department of Computer Science at the University of Warwick. More information can be found on his homepage (www.lfhsgre.org).

Human activity recognition (HAR) is a compelling topic in the fields of ubiquitous computing, with numerous applications including human behaviour understanding, smart healthcare, human-computer-interaction (HCI), and more. Among various sensing technologies, Radio Frequency (RF) signals such as WiFi can be applied in a less intrusive manner, offering significant potential in smart home environments. However, due to multipath propagation, WiFi signals often contain a high level of noise from physical environments that need to suppressed before developing HAR models.