|
Big HypothesesPrinciple Investigator: Simon Maskelllihttps://www.liverpool.ac.uk/electrical-engineering-and-electronics/staff/simon-maskell/ nk Project DescriptionWe are developing scalable solutions to numerical Bayesian inference by using a judiciously chosen configuration of an SMC sampler (a readily parallelised alternative to MCMC) with a focus on providing game-changing speed-ups to problems that have broad relevance across many industries and scientific endeavours.
|
|
AI techniques for automated diagnosis of eye diseasesPrinciple Investigator: Yalin Zhenghttps://www.liverpool.ac.uk/ageing-and-chronic-disease/staff/yalin-zheng/ Project DescriptionDeep learning (DL) has transformed image recognition and analysis by enabling unprecedented performance improvements in computer vision tasks in recent years. These rapid advancements enabled the development of automated, accurate, accessible, and cost-effective image recognition AI for medical diagnostics. In this project, we will explore the potential of deep learning techniques to develop reliable and accurate tools to better support early detection and management of eye diseases. Eye diseases represent a global health challenge. It is expected that these new tools will be introduced into clinical practice and benefit millions of patients' quality of life and reduce the cost of healthcare provision.
|
|
Code generation for distributed model parallelismPrinciple Investigator: Navjot Kukrejahttps://www.liverpool.ac.uk/computer-science/staff/navjot-kukreja/ Project DescriptionThis project uses code generation for distributed model-parallel training of deep neural networks. This is specific to scientific applications, in particular, those that deal with 3D data. There is a chicken and egg problem in this space where not enough progress is being made, because the tools that could handle the volumes of data that these applications need do not exist, and the tools do not exist because there are not enough users. Here we leverage the power of code generation to generate highly optimised code for distributed model-parallel training at scale.
|
|
Deep Federated Learning for Autonomous DrivingPrinciple Investigator: Anh Nguyenhttps://www.csc.liv.ac.uk/~anguyen/ Project DescriptionAutonomous driving is an emerging field that has the potential to transform the way humans travel. The most recent approaches for autonomous driving are based on machine learning, particularly deep learning techniques that require large-scale datasets. While collecting data will help to develop more accurate autonomous driving solutions, it strongly violates user privacy as personal and sensitive data are shared with centralized servers. A promising solution for this problem is federated learning. Federated learning allows us to train a model in a decentralized manner to maintain users' privacy. This project will develop new methods to train deep networks in a decentralized way for autonomous driving by designing new deep federated learning topologies and developing asynchronous optimization methods to ease model convergence when training large-scale deep models.
|
|
High-throughput simulation of polymersPrinciple Investigator: Hesam Makkihttps://www.liverpool.ac.uk/chemistry/staff/hesam-makki/ Project DescriptionVirtual material screening has shown great potential in small molecules and crystal structures discovery; however, polymer design has not benefited from digital tools as much, due to the many obstacles in the path of accelerating polymer modelling. Polymer simulation has traditionally focused on the study of benchmark systems. Thus, the community is not actively exploring the potential of large homogeneous datasets, enabled by robotics synthesis and exploration. Therefore, a new standard in polymer modelling needs to be established to enable expediting the simulation process. This project aims to enable high-throughput polymer modelling and unlock the digital discovery of polymers through the development of novel computational approaches.
|
|
PolyhymniaPrinciple Investigator: Eduardo Coutinhohttps://www.liverpool.ac.uk/music/staff/eduardo-coutinho/ Project DescriptionResearch shows that music can be used to improve mental health. It can help people relax, distract them from worries, and even connect with others. Studies have also shown that listening to music over time can actually lessen depression symptoms, even without a doctor's help. It can even work better than some other treatments on its own! However, there is a caveat! Not all music will make us feel better. The wrong music can make our depressing thoughts get stuck in a negative thought loop, or make it seem worthless to make an effort. Therefore, using music to improve mood requires some skill and know-how, especially when one is feeling low. In this project, we are creating an experimental app called Polyhymnia Mood that helps people create personalized playlists to improve their mood in effective and healthy ways. Early tests showed it worked well, and people using the app saw an improvement in their mood and depression symptoms. Now, we want to make it even more effective.
|