Introduction
This PhD project focuses on leveraging Neurosymbolic Machine Learning (ML) to drive advancements in Sustainable AI, particularly by optimising the efficiency of AI systems to reduce energy consumption.
Neurosymbolic ML, which combines the interpretability of symbolic AI with the flexibility of subsymbolic techniques like deep learning, is a promising approach to achieving sustainability in AI. Symbolic reasoning, with lower computational demands, can significantly reduce the resource consumption of AI models, making them more eco-friendly without sacrificing accuracy or adaptability.
The focus of this PhD project will be on using neurosymbolic ML methods to analyse Distributed Fibre Optic Sensing (DFOS) data. DFOS technology is capable of providing highly granular data on environmental vibrations, structural integrity, and other signals critical for monitoring urban infrastructure and natural environments. One of the key challenges of this technology is the vast density of data, which needs massive storage solutions, as well as computational power to analyse it. The student will explore whether neurosymbolic methods can be leveraged to reduce the data storage and computational resources necessary to work with this type of data.
This research will be tightly aligned with several research projects across the University and the National Oceanography Centre (NOC), using DFOS collected in different environments. In particular, the student will benefit from DFOS data collected in Southampton and London, one of the largest-to-date datasets of this kind in urban environment and first in the UK. Moreover, the student will also have access to data held at NOC, collected through different campaigns in marine environments, which could be applied for research on AI and environment. The availability of these diverse data sources will enable the investigation of noise-tolerant sustainable ML models capable of operating in multiparameter environments, such as those influenced by urban infrastructure or marine ecosystems. Underlying this is the SOUNDSCALE project (https://soundscale.ac.uk/), with collaborators from different fields, from social sciences to humanities, to collectively analyse the impacts of DFOS data in urban settings. Students with interest in impact and interdisciplinary research will find opportunities for career progression in these areas.
Challenges
The main research challenge lies in developing event detection systems that can operate effectively in complex, noisy environments while maintaining low energy consumption.
Traditional deep learning methods, though powerful, require significant computational resources, especially when applied to large datasets like those generated by DFOS platforms. Neurosymbolic ML, by contrast, offers a more computationally efficient approach, enabling the development of models that can detect and classify events with reduced energy demands.
Supervisory Team
You will be supervised by a team of interdisciplinary researchers in machine learning, signal processing and distributed systems, and will have the opportunity to collaborate with industry partners to further your research. You will join the School of Electronics and Computer Science in collaboration with the National Oceanography Centre (NOC), and will have professional development opportunities through the Alan Turing Institute, the UKRI TAS Hub, Responsible AI UK (Rai UK) and access to Future Worlds to explore commercialisation for your research.
Themes
While the primary theme of this project is Sustainable AI, it also has broader implications for other SustAI CDT themes. DFOS data can be applied to sectors such as Transportation and Logistics, where real-time monitoring and optimisation can reduce resource consumption and enhance sustainability. Moreover, it could be applied for environmental research, to assess the effects of climate change in coastal cities, detection of extreme weather events, and so on.