Recent Artificial Intelligence Research Centre projects
Find out more about what we're working on at AIRC as well as projects that we've recently completed.
- ASGARD: Analysis System for Gathered Raw Data
DESIREE: Decision Support and Information Management System for Breast Cancer
This project looked at medical image analysis, data and knowledge modelling, case-based reasoning to transform decision making for breast cancer diagnosis and treatment.
VIAD: Fast video-based anomaly detection on railway tracks
The VIAD project focuses on the problem of detecting anomaly on high-speed rail tracks based on video streams from train mounted cameras. The tracks are expected to be clear of any objects, moving or stationary, within a certain peripheral range.
The tracks are also expected to be in good shape. If any of these expected conditions are violated, the scene can be deemed abnormal and an alarm may be raised.
The aim of this project is to design a novel anomaly measure to enable a novel framework for environmental anomaly detection based on video streams from train mounted cameras, and evaluate on real video data through the China partners.
VIPIRS: Virus Identification via Portable InfraRed Spectroscopy
Spectroscopic techniques such as infra-red, Raman, and mass spectrometry have long been used to identify chemical compounds and biological species, including bacteria and viruses, usually in specialised lab conditions with high performance instrumentation.
Virus identification in realistic clinical/field environments, using low cost instrumentation, is appealing as it can be widely deployed and so is very suitable for diagnosis, prevention and management in pandemics such as COVID-19. However, low cost instrumentation produces poorly-resolved spectra with added noise.
In this project we will study virus spectral characteristics and investigate how to analyse low quality spectral data. We will then build spectral data-based virus detection models. We aim to develop a spectrometer-fronted, cloud-based system for in-situ virus detection.
The system can record spectra from patient samples and return a positive/negative diagnosis quickly, based on detection models running on a cloud-based service. The system will be validated in realistic environments in collaboration with Northern Ireland Regional Virology Lab.
MVSE – Multimodal Video Search by Examples
How to effectively and efficiently search for content from large video archives such as BBC TV programmes is a significant challenge. Search is typically done via keyword queries using pre-defined metadata such as titles, tags and viewer's notes. However, it is difficult to use keywords to search for specific moments in a video where a particular speaker talks about a specific topic at a particular location.
Video search by examples is a desirable approach for this scenario as it allows search for content by one or more examples of the interested content without having to specify interest in keyword. However, video search by examples is notoriously challenging, and its performance is still poor. To improve search performance, multiple modalities should be considered – image, sound, voice and text, as each modality provides a separate search cue so multiple cues should identify more relevant content.
This is multimodal video search by examples (MVSE). In this project we will study efficient, effective, scalable and robust MVSE where video archives are large, historical and dynamic; and the modalities are person (face or voice), context, and topic.
The aim is to develop a framework for MVSE and validate it through the development of a prototype search tool. Such a search tool will be useful for organisations such as the BBC and British Library, who maintain large collections of video archives and want to provide a search tool for their own staff as well as for the public.
It will also be useful for companies such as Youtube who host videos from the public and want to enable video search by examples. We will address key challenges in video segmentation, content representation, hashing, ranking and fusion.
CESAR: Conjunctive Explanations
This project explores the nature of explanatory and scientific reasoning. In particular, it seeks to address the question: how can two (or more) or more explanatory hypotheses work together as a ‘conjunctive explanation’ rather than as distinct, competing explanations? The project involves research on Bayesian measures of explanatory goodness and simulations of abductive inference. It also explores the relevance of these issues to explanations in science and religion and is carried out in collaboration with researchers in philosophy of science (University of Utah) and history of science and religion (Queen’s University Belfast).
- Dates: Oct 2018 – June 2021
- Researchers: David Glass (PI), Mark McCartney, Jonah Schupbach, Diarmid Finnegan, David Livingstone
- Funded by: John Templeton Foundation
Projects recently completed
Explaining and Explaining Away
In this project we explored the concept of explaining away in order to show that it provides a fruitful way to understand various aspects of the interaction between science and religion.
The main research activities were the development of the theory of explaining away, the application of this work to a science and religion context, and the development and implementation of models for simulating how social groups update their beliefs in contexts where both scientific and religious explanations are present.
Outputs will be in the form of academic papers, presentations at conferences, popular articles, and in the organization of an interdisciplinary workshop for academics and a training event for ministers and laypersons.
This project will make a difference for academics by showing that the concept of explaining away can shed new light on some discussions in the field of science and religion and that formal approaches have a powerful and useful role to play in understanding the science and religion dynamic. It also aims to raise awareness to a lay audience that scientific explanations need not undermine religious explanations, but that they can be mutually supportive.
The research has the potential to generate novel avenues in the dialogue between science and religion by providing a new way in which apparent conflicts can be resolved and by developing computational models of belief change in a social group.
Given that one aspect of the project’s innovation comes from its interdisciplinary nature, drawing on research in the fields of artificial intelligence, formal epistemology and computer simulations of complex social networks, it is hoped that it will also have an impact beyond the field of science and religion.
DASGED: Detecting Anomalies from Satellite and Terrestrial Electromagnetic Data Sources by Big Data Analytics
The evolutionary process of earthquakes is rich of complex features, from stochastic to chaotic or pseudo-periodic dynamics, often depending on the different geo-tectonic regime and focal mechanisms.
A wide range of fundamental research in the lithosphere, such as from fault rupture mechanics to seismic wave propagation, has been undertaken on different regions in past decades.
Many case studies show there are some seismic anomalies before earthquakes and associated phenomena after earthquakes in respect of ground deformation, active faults (slip rates and geometry), tectonic stress fields and geomagnetic fields, but systematic understanding of preparation process of earthquakes and their seismic cycles is very limited to date.
There is no pragmatic approach to predicting earthquakes. Earth observation (EO) building on satellites provides the enormous capability of observing regional and global areas of our planet. It records a wide range and huge volumes of spatial-temporal measurements, which capture a variety of activities occurring on the earth and in its interior. Meanwhile, EO provides new possibilities for scientists to investigate the earthquake behaviour by using a broad range of abnormal phenomena reflected in ionosphere and ground displacement activities observed from space and improve understanding of preparation process of earthquakes from the world scale and the perspectives of coupling ionosphere with lithosphere.
The project is aimed to investigate spaceborne DEMETER data, SWARM, SAR and other satellite data for detecting a range of precursors reflected in ionosphere and lithosphere, before large earthquakes and to assess correlations between the selected precursors and seismic events. The specific objectives are to
- evaluate HJ-1B, CBERS-01&02, CBERS-03, HY-1 & 2, ENVISAT/ASAR, etc. satellites for monitoring deformation and thermal infrared spectrum, and DEMETER, SWARM and CSBS for electromagnetic emission; *improve our knowledge and understanding on ionospheric perturbations and abnormal changes in electromagnetic emission intensity and electron density using DEMETER, SWARM, CSBS;
- develop effective methodologies based on computational intelligence (machine learning and data mining), statistical methods and existing methods developed for handling ground deformation and thermal infrared spectrum to extract anomalies;
- define a uniform entropy measure to transform the diverse anomalies to a uniform representation without loss of information content and precision;
- perform information fusion on unified anomalies and examine the divergence among combined results and select convergent anomalies for further analyses;
- analyze correlations and coherences between the anomalies and seismic cycles and;
- develop a uncertainty model based on information fusion theory, Bayesian probability and Theory of Belief Functions to provide an approach to estimating probabilities of potential seismic activities in a given region.
MEWODM: Monitoring and Early Warning of Epidemic Diseases by Earth Observation Data Mining
Many of epidemic and infectious diseases are closely related to natural environment for their living, breeding and transmission of pathogens, especially vector-borne diseases (e.g. schistosomiasis) that rely heavily on their vectors.
The project is aimed at developing a viable approach for discovering and monitoring diseases, which will be used to prevent and control vector-borne diseases.
SMS: Sentiment Mining for Security
Increase competitiveness and growth by developing unique sentiment-mining techniques, utilising cloud-based parallel/multi-core processing resources, to increase data-scaling across multiple data-centres, databases and processing engines for detecting security related events.
- SAVASA: Standards based Approach to Video Archive Search and Analysis
DEEPFLOW: Extraction of argument structures from text to support innovation and decision
DeepFlow seeks to address some of the key problems related to understanding and managing interrelated design documents as collated on a given manufacturing project in the Aerospace, Defence and Automotive industries.
Design specification documents consist of large sections of unstructured textual data, many of which over the duration of the project are subject to revision based on a expert-led discourse over the strengths and weaknesses of the various design components.
The project aims to form a unique combination of research into natural language processing and argumentation theory, to develop a project summarization tool that assists engineers to comprehend the topical content and various discourse themes within a given project. Such a tool is of enormous benefit to the industry as a design project may typically contain many thousands documents.
The tool will have the added benefit of allowing different projects to be compared and contrasted in terms of their underlying content.
This provides a significant progression over current document management applications currently in use. Such systems provide only limited text analysis mechanisms whereas DeepFlow will allow designers to be able to discover the gist of a project and quickly re-evaluate the design elements of existing specifications, when considering a new project.