AI Hopes and Fears
7 November 2023
We are delighted to announce the 11 projects funded through this years’ Accelerate Science and Cambridge Centre for Data Driven Discovery (C2D3) funding call which took place this summer.
This year’s awardees will all explore how we can successfully deploy AI to tackle real-world and research challenges. Projects range from workshops bringing together researchers at the nexus of artificial intelligence and early cancer detection, a research and training initiative for investigative journalists to the development of a monitoring system combining computer vision and AI in neonatal intensive care.
Successful projects will deliver results over the coming year and we will share updates through the Accelerate Programme blog. The wide range of projects supported through this call will catalyse further discoveries to accelerate scientific progress and create AI tools that are capable of delivering benefits for science and society.
The funded projects are:
Machine Learning for Investigations in the Public Interest Dr Anne Alexander, Director of Learning, Cambridge Digital Humanities
Currently there are few opportunities open to journalists and NGOs to acquire skills and practical knowledge about how to deploy ML as part of investigations in the public interest. Through this project, Anne and colleagues in Cambridge and at the Pulitzer Centre will be addressing this gap by carrying out and documenting a collaborative investigation into illegal mining in the Amazon basin and creating a network of practitioners to disseminate good practice and train others.
From Theory to Treatment: Advancing Diffusion Models for Medical Diagnosis Angelica Aviles-Rivero, Senior Research Associate, Department of Applied Mathematics & Theoretical Physics
Imaging is a key tool in medical diagnosis but images and be ‘noisy’, containing multiple sources of information and are sometimes incomplete. The multidisciplinary project team will pioneer novel adaptive sampling techniques to streamline the training process of diffusion models used in image analysis. This innovation will significantly reduce computation resources, rendering diffusion models more practical for real-world applications, including medical imaging diagnosis.
PaleoStats workshop: AI and Statistical Innovations for Paleoecological Research Marco A. Aquino-López, Research Associate, Department of Geography
Environmental reconstructions serve as crucial tools for understanding past climatic and ecological conditions. However, uncertainties in the models compromise our ability to understand past environmental changes and project future scenarios. The project team will bring together experts in paleoecology, climate science, statistics, and AI to collaborate with the goal to develop an AI-driven tool that can more efficiently propagate age-related uncertainties.
Preclinical studies of novel antibiotics discovered by AI Sergio Bacallado de Lara, Associate Professor, Department of Pure Mathematics and Mathematical Statistics
Antimicrobial resistance poses a threat to global health. Despite the urgency for novel therapeutics, the number of new drug candidates remains low. Working with collaborators at the University of Cambridge and the University of Warwick, the team will adopt an interdisciplinary transfer-learning approach to antibiotic discovery. Through this project they will conduct pre-clinical testing of molecules identified as having indicators of antibiotic activity and explore translation pathways to the clinic.
Machine Learning and AI for Hard-To-Treat Cancers: Datasets, Pipelines and Clinical Implementation Symposium Dr Mireia Crispin-Ortuzar, Assistant Professor, Dept of Oncology and co-lead, Institute for Integrated Cancer Medicine
Cancer survival has doubled in the past 40 years, but survival rates of some cancers such as lung, brain, or ovarian have hardly moved. Recent developments in AI have unlocked possibilities to advance our understanding of cancer’s underlying biology. The project team will host an International Symposium on AI for Cancer Data Integration, bringing together clinicians, patients, AI researchers and industry partners, showcasing existing cutting-edge projects, and providing a forum for continued collaboration and interaction.
Meerkat Neonatal Monitoring – Embedded Device Development Alex Grafton, Research Associate, Department of Engineering
Meerkat, developed in a collaboration between the Departments of Engineering and Paediatrics, provides a second set of eyes for nurses caring for babies in Neonatal Intensive Care Units (NICU) with insight from machine learning technology, ensuring clinically significant events are noted and acted upon. The project will use real-world data to optimise models for edge computing on embedded devices in the NICU environment.
Developing robust cancer early detection systems: Navigating distribution shifts across cohorts and time Samantha Ip, Research Associate, Department of Public Health and Primary Care
Cancer diagnosis often occurs at later stages, resulting in poor prognoses. To address this issue, considerable effort has been invested in developing early cancer detection models. A key challenge is distribution shifts, where target distribution deviates from training data. The project team will bring together experts for a 2 day event to develop modelling approaches and mechanisms to accommodate distribution shifts.
Harnessing AI to unlock natural history collections for zoonotic surveillance Maya Juman, PhD Student, Department of Veterinary Medicine
Zoonotic spillover, the transmission of pathogens from animals to humans, is incredibly costly in terms of human mortality and economic impact. Understanding where pathogens originate is critical to prevent spillover. The project team will develop a machine learning algorithm to study fruit bats known to host paramyxoviruses. The team will work with natural history museum collections to empirically test their model and build risk maps to inform future viral surveillance.
An artificial intelligence-derived decision support tool to predict outcomes after recanalisation treatments in stroke patients James Rudd, Professor of Cardiovascular Medicine, Victor Phillip Dahdaleh Heart & Lung Research Institute
Stroke is a leading cause of death and disability in the UK. Despite highly effective therapies, 1 in 4 patients in trials have poor long-term outcomes. The team will develop ML methodology to predict clinical outcomes after stroke treatment by integrating both clinical and imaging data. Findings will be validated in an external cohort of stroke patients across East Anglia in collaboration with the East of England Telestroke network.
Predicting decisions of the Court of Appeal using AI Felix Steffek, Professor of Law, Faculty of Law
This project will use machine learning and natural language processing to predict the outcome of court litigation. The team will leverage state-of-the-art machine learning approaches to predict general and detailed case outcomes, and will investigate the ability of systems to provide clear and interpretable explanations for these predictions.
Embodied Artificial Intelligence and Bio-inspired Soft Robotics (Embodied AI and Bi-SoRo) Yue Xie, Marie Sklodowska-Curie Future Roads Fellow, Bio-Inspired Robotics Laboratory, Department of Engineering
The project team will host a workshop for experts from AI, robotics, bioengineering and related fields. This interdisciplinary dialogue will enable participants to gain a comprehensive understanding of the challenges and opportunities presented by AI in robotics. Leveraging the collective expertise of the participants, the project will lead to the design innovative solutions bridging the gap between AI algorithms and embodied agents.