A5048801815- Topic Modeling
- Semantic Web and Ontologies
- Natural Language Processing Techniques
- COVID-19 epidemiological studies
- Bayesian Modeling and Causal Inference
- Vaccine Coverage and Hesitancy
- Explainable Artificial Intelligence (XAI)
- Advanced Graph Neural Networks
- Speech and dialogue systems
- Service-Oriented Architecture and Web Services
- Misinformation and Its Impacts
- Multimodal Machine Learning Applications
- Advanced Database Systems and Queries
- Logic, Reasoning, and Knowledge
- Data Mining Algorithms and Applications
- Data Quality and Management
- Data-Driven Disease Surveillance
- COVID-19 diagnosis using AI
- Data Stream Mining Techniques
- Complex Network Analysis Techniques
- Advanced Clustering Algorithms Research
- Scientific Computing and Data Management
- SARS-CoV-2 and COVID-19 Research
- COVID-19 Digital Contact Tracing
- Biomedical Text Mining and Ontologies
University of Edinburgh
2016-2023
Large Language Models (LLMs) excel in natural language tasks but still face challenges Question Answering (QA) requiring complex, multi-step reasoning. We outline the types of reasoning required some these tasks, and reframe them terms meta-level (akin to high-level strategic or planning) object-level (embodied lower-level such as mathematical reasoning). Franklin, a novel dataset with requirements meta- reasoning, is introduced used along three other datasets evaluate four LLMs at question...
Infectious disease surveillance is difficult in many low- and middle-income countries. Information market (IM)-based participatory a crowdsourcing method that encourages individuals to actively report health symptoms observed trends by trading web-based virtual "stocks" with payoffs tied future event.
Logs record system events and status, which help developers administrators diagnose run time errors, monitor running status mine operation patterns [13, 23]. However, logs are complex weakly linked, making it difficult to the causes of failures. While recent studies on log knowledge extraction focus lifting entities from messages for enriching a background graph (BKG), they do not involve reasoning inferring implicit relations nor guarantee that learned streams is consistent with knowledge....
Report 3, published 19 July 2021. Fully open-access.We conducted a nationally representative online survey in Ghana (N = 1295) throughout June 2021.In our analyses, we operationalised vaccine hesitancy as respondents who answered ‘no’ and ‘I don’t know’ to the question: “When COVID-19 becomes available you, would you like get vaccinated?” Some top-level findings share - willingness vaccinate dropped from 82% March, 71% 2021 - Therefore, phrase another way, there was an observed...
This policy brief analyses data collected in Togo shortly before vaccines were officially approved. We conducted a nationally representative telephone survey (N = 1,558) throughout December 2020, prior to COVID-19 arriving Togo. In our analyses, we operationalised vaccine hesitancy as respondents who answered ‘no’ and ‘I don’t know’ the question: “When becomes available you, would you like get vaccinated?” Rates of was fairly high, with 67.7% Togolese said that they willing be vaccinated...
An important aspect of artificial intelligence (AI) is the ability to reason in a step-by-step "algorithmic" manner that can be inspected and verified for its correctness. This especially domain question answering (QA). We argue challenge algorithmic reasoning QA effectively tackled with "systems" approach AI which features hybrid use symbolic sub-symbolic methods including deep neural networks. Additionally, we while network models end-to-end training pipelines perform well narrow...
Both symbolic and sub-symbolic AI have their limi-tations, but combination can be more than the sum of parts. For instance, statistical machine learning has been hugely successful at classification decision-making tasks, not so good deliberative systematic reasoning nor explanation. We argue that by combining into hybrid systems, whole will its To illustrate potential system, we describe FRANK query answering system. infers new knowledge from diverse immense sources on Web, using a both...
Recent developments in support for constructing knowledge graphs have led to a rapid rise their creation both on the Web and within organisations. Added existing sources of data, including relational databases, APIs, etc., there is strong demand techniques query these diverse knowledge. While formal languages, such as SPARQL, exist querying some graphs, users are required know which they need unique resource identifiers resources need. Although alternative neural information retrieval embed...
Knowledge Graph (KG) Construction is the prerequisite for all other KG research and applications. Researchers engineers have proposed various approaches to build KGs their use cases. However, how can we know whether our constructed good or bad? Is it correct complete? consistent robust? In this paper, propose a method called LP-Measure assess quality of via link prediction tasks, without using gold standard human labour. Though theoretically, only consistency redundancy, instead more...
<sec> <title>BACKGROUND</title> Infectious disease surveillance is difficult in many low- and middle-income countries. Information market (IM)–based participatory a crowdsourcing method that encourages individuals to actively report health symptoms observed trends by trading web-based virtual “stocks” with payoffs tied future event. </sec> <title>OBJECTIVE</title> This study aims assess the feasibility acceptability of tailored IM system monitor population-level COVID-19 outcomes Accra,...
We present a study into the ability of paraphrase generation methods to increase variety natural language questions that FRANK Question Answering system can answer. first evaluate on LC-QuAD 2.0 dataset using both automatic metrics and human judgement, discuss their correlation. Error analysis is also performed manual approaches, we how evaluation affected by data points which contain error. then simulate an implementation best performing method (an English-French backtranslation) in order...
When the estimated probabilities do not match relative frequencies, we say these are uncalibrated [39], which may cause incorrect decision making, and is particularly undesired in high-stakes tasks [45]. Knowledge Graph embedding models reported to produce [36], e.g., for all triples predicted with probability 0.9, percentage of them being truly correct . In this article, take a closer look at problem. First, confirmed issue that typical KG Embedding uncalibrated. Then, show how...