DiPET - Distributed Stream processing on Fog and Edge Systems via Transprecise Computing
The DiPET project investigates models and techniques that enable distributed stream processing applications to seamlessly span and redistribute across fog and edge computing systems. The goal is to utilize devices dispersed through the network that are geographically closer to users to reduce network latency and to increase the available network bandwidth. However, the network that user devices are connected to is dynamic. For example, mobile devices connect to different base stations as they roam, and fog devices may be intermittently unavailable for computing. In order to maximally leverage the heterogeneous compute and network resources present in these dynamic networks, the DiPET project pursues a bold approach based on transprecise computing. Transprecise computing states that computation need not always be exact and proposes a disciplined trade-off of precision against accuracy, which impacts on computational effort, energy efficiency, memory usage and communication bandwidth and latency. Transprecise computing allows to dynamically adapt the precision of computation depending on the context and available resources. This creates new dimensions to the problem of scheduling distributed stream applications in fog and edge computing environments and will lead to schedules with superior performance, energy efficiency and user experience. The DiPET project will demonstrate the feasibility of this unique approach by developing a transprecise stream processing application framework and transprecision-aware middleware. Use cases in video analytics and network intrusion detection will guide the research and underpin technology demonstrators.
Project PI: Prof Hans Vandierendonck
Universitat Politècnica de Catalunya - Spain
Institut de Recherche en Informatique et Systèmes Aléatoires - France
Foundation for Research and Technology - Hellas - Greece
West university of Timisoara - Romania
- UniServer - A €4.8M project funded by Horizon Europe coordinated by Queen's University Belfast
UniServer stands for, “A Universal Micro-Server Ecosystem by Exceeding the Energy and Performance Scaling Boundaries”. Manufacturing variance in semi-conductor chips means that micro-processors and memory chips produced by the same process may be marketed and sold at vastly different prices. This occurs because some chips fail high-frequency tests and are then sold off as lower powered devices. The UniServer project builds system software that pushed chips to their limits and then keeps that boundary information on a per device basis. System software can then make use of this boundary information to make more informed decisions on how to bootstrap devices and maximise the true compute power that they offer. By leveraging the previously conservative voltage/frequency margins adopted in commercial processors and memory chips, UniServer enhances performance with new margin/fault-aware runtime and resource management policies. This strategy delivers much more power efficient, low-cost micro-servers aimed at the emerging Big Data and IoT marketplaces.
ARM, UK; Applied Micro Circuits Corporation, Germany; IBM, Ireland; WORLDSENSING SL; MERITORIUS AUDIT LTD; SPARSITY SL
Project Members: Dr Georgios Karakonstantis, Dr Charles Gillan
Output: Book titled Computing at the EDGE: New Challenges for Service Provision https://books.google.ie/books/about/Computing_at_the_EDGE.html?id=K4RXzgEACAAJ&redir_esc=y
- Clinical Narrative Analytics: Computational Linguistics to Predict Dementia Diagnosis from Descriptive Clinical Text
This was a 12 month project with the Health and Social Care Board of Northern Ireland The application of artificial intelligence (AI) algorithms and machine learning (ML) has the potential to revolutionize health care, supporting clinicians, providers, and policymakers to plan or implement interventions, detect disease quicker, support therapeutic decision-making, outcome prediction, and increased personalized medicine.
- To increase the utility of machine learning in healthcare analytics new algorithms and analytical models are required to interpret and classify unstructured text within clinical documentation.
- By digitising and pre-processing 484 clinical letters this project sought to develop clinical notes analytical software that uses automated tools for analysing clinical letters.
The overall goal of the project was to build a computational system that can analyse and extract meaning from free-form clinical narrative text, and to investigate the potential applications to dementia care and diagnosis.
There were four key objectives of the project, organised around the 4 sprints:
- Acquire a dataset of clinical letters that could be used in the project. This objective involved the de-identification and optical character recognition of clinical letters relating to patients with various dementia-related conditions (e.g. Alzheimer's Disease, MCI, and healthy individuals). In total, with our collaborators at 3 clinics across Northern Ireland, our goal was to obtain approximately 1000 clinical letters.
- Build Natural Language Processing AI tools to represent the word and sentence meanings in the clinical text. We aimed to use recent Neural Network based NLP tools for this task. These networks represent the meaning of words and sentences as distributed patterns of activity within the network. As part of this objective, we investigate and implement data augmentation techniques to efficiently use all sentences for model training.
- Predictive modelling of patient diagnosis: using the systems described in (b) above, our objective was to build models through supervised training to predict patient diagnosis. We aimed to investigate many different network architectures in this step, and we also wished to compare results with more traditional supervised learning techniques such as random forests and logistic regression.
- Downstream applications of the system: we aimed to implement a prototype clinical decision support application and statistical analysis of other similar clinical data.
The project produced a proven, externally validated, state-of-the-art AI system for representing the meaning of sentences in clinical text. The project team has had a focus on identifying commercial applications and has identified a potential use case in identifying subjects for clinical trials and / or wider longitudinal research studies.
From a technical data analytics perspective this project has delivered a significant outcome, and as such it has also raised the profile of dementia analytics capability (and Computational Linguistics capability more generally) in Northern Ireland on an international stage.
Project PI: Dr Barry Deveruex
Project Partners: Health and Social Care Northern Ireland
CITI-GENS is a Horizon2020 funded Marie Skłodowska-Curie doctoral training programme that supports 20 interdisciplinary PhD students at Queen’s University Belfast. PhD students funded within Marie Skłodowska-Curie programmes have the title of Early Stage Researcher (ESR). All researchers will have commenced by April 2021.
Project Title: Iterative Approximate Analysis Of Graph-Structured Data For Precision Medicine
The advent of technology and the growth of computers prompted researchers to focus on problems that are NP-hard in nature. With the growth of technology, graph-structured data also increases exponentially each year. Finding the maximum weighted clique from such a vast amount of graph-structured data is a challenging task due to its complexity and huge size. Although several attempts have been made to solve the maximum weighted clique problem in huge graphs, there is still much opportunity for lowering the execution time necessary to find a satisfactory solution. We intend to provide a unique solution to solve the maximum weighted clique problem from the weighted undirected graphs with a reasonable trade-off between execution time and quality solution.
Industry partner: IBM Research Europe – Zürich
IBM will provide industrially relevant context on knowledge extraction from graph-structured data. They have extensive experience in this area by building scalable software systems for the analysis of massive-scale graph data. They will moreover provide access to relevant datasets.
Project Title: Fair AI for the Public Sector: Advancing the Technological Frontier
Machine learning (ML), the discipline that forms the contemporary mainstream of AI research, targets to develop algorithms designed to leverage patterns within datasets to address specific tasks and challenges. By way of example, a simple association (rule) mining algorithm might identify insights from shopping patterns to inform decisions on product placement in stores, or to send marketing offers to customers. Left to be guided with objectives such as maximization of (predictive) accuracy or its economic manifestations such as advertising revenue, conventional ML algorithms can leverage the existence of benign patterns of user behaviour present in the dataset, such as frequent co-buying of milk and bread.
However, the data often contains other kinds of patterns as well, those which would be considered less benign and instead more reflective of societal biases that legitimate buying choices. ML algorithms have no way to differentiate between desirable and undesirable patterns, being only guided by maximization of accuracy. Both kinds of patterns would be leveraged in the insights they generate given that they are both likely useful. Thus, ignoring undesirable patterns to make fairer decisions incurs a cost in terms of conventional measures such as accuracy and advertising revenue.
These concerns are supremely important to the issue of ML algorithms within the public sector, where fairness is an internationally-understood value in citizen-state interactions and the elimination of bias a legal and organisational objective of public sector organisations.
This project focuses on the development of fair ML algorithms for unsupervised learning tasks such as clustering, retrieval, representation learning and anomaly detection. In practice, this project will involve developing an understanding of the principles of justice and fairness from political philosophy and translating them into mathematical formulations upon which scenario-specific bespoke fair ML algorithms would be built. The project will have considerable potential for the increasing number of public sector organisations that rely on ML to conduct transaction and other processes within government.
Project PI: Dr Deepak Padmanabhan
Industry Partner: IBM Research - India