Patrick J. Wall
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Patrick J. Wall has a total of 61 co-authors. Here are the top co-authors by number of shared publications:
- Richard Heeks
University of Manchester
5 shared publications - David Lewis
ADAPT Centre, Trinity College Dublin, Dublin, Ireland
4 shared publications - Deepak Saxena
Indian Institute of Technology, Jodhpur, India
4 shared publications - Suzana Brown
SUNY Korea
4 shared publications - Lucy Hederman
School of Computer Science and Statistics, Trinity College Dublin, Dublin, Ireland
3 shared publications - Cathy Roche
Trinity College Dublin
2 shared publications
Latest Publications
- Open Access
- 21 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 8091
Confidence-Based Knowledge Distillation to Reduce Training Costs and Carbon Footprint for Low-Resource Neural Machine Translation
The transformer-based deep learning approach represents the current state-of-the-art in machine translation (MT) research. Large-scale pretrained transformer models produce state-of-the-art performance across a wide range of MT tasks for many languages. However, such deep neural network (NN) models are often data-, compute-, space-, power-, and energy-hungry, typically requiring powerful GPUs or large-scale clusters to train and deploy. As a result, they are often regarded as “non-green” and “unsustainable” technologies. Distilling knowledge from large deep NN models (teachers) to smaller NN models (students) is a widely adopted sustainable development approach in MT as well as in broader areas of natural language processing (NLP), including speech, and image processing. However, distilling large pretrained models presents several challenges. First, increased training time and cost that scales with the volume of data used for training a student model. This could pose a challenge for translation service providers (TSPs), as they may have limited budgets for training. Moreover, CO2 emissions generated during model training are typically proportional to the amount of data used, contributing to environmental harm. Second, when querying teacher models, including encoder–decoder models such as NLLB, the translations they produce for low-resource languages may be noisy or of low quality. This can undermine sequence-level knowledge distillation (SKD), as student models may inherit and reinforce errors from inaccurate labels. In this study, the teacher model’s confidence estimation is employed to filter those instances from the distilled training data for which the teacher exhibits low confidence. We tested our methods on a low-resource Urdu-to-English translation task operating within a constrained training budget in an industrial translation setting. Our findings show that confidence estimation-based filtering can significantly reduce the cost and CO2 emissions associated with training a student model without drop in translation quality, making it a practical and environmentally sustainable solution for the TSPs.
- 21 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 8091
Open AccessConfidence-Based Knowledge Distillation to Reduce Training Costs and Carbon Footprint for Low-Resource Neural Machine Translation
The transformer-based deep learning approach represents the current state-of-the-art in machine translation (MT) research. Large-scale pretrained transformer models produce state-of-the-art performance across a wide range of MT tasks for many languages. However, such deep neural network (NN) models are often data-, compute-, space-, power-, and energy-hungry, typically requiring powerful GPUs or large-scale clusters to train and deploy. As a result, they are often regarded as “non-green” and “unsustainable” technologies. Distilling knowledge from large deep NN models (teachers) to smaller NN models (students) is a widely adopted sustainable development approach in MT as well as in broader areas of natural language processing (NLP), including speech, and image processing. However, distilling large pretrained models presents several challenges. First, increased training time and cost that scales with the volume of data used for training a student model. This could pose a challenge for translation service providers (TSPs), as they may have limited budgets for training. Moreover, CO2 emissions generated during model training are typically proportional to the amount of data used, contributing to environmental harm. Second, when querying teacher models, including encoder–decoder models such as NLLB, the translations they produce for low-resource languages may be noisy or of low quality. This can undermine sequence-level knowledge distillation (SKD), as student models may inherit and reinforce errors from inaccurate labels. In this study, the teacher model’s confidence estimation is employed to filter those instances from the distilled training data for which the teacher exhibits low confidence. We tested our methods on a low-resource Urdu-to-English translation task operating within a constrained training budget in an industrial translation setting. Our findings show that confidence estimation-based filtering can significantly reduce the cost and CO2 emissions associated with training a student model without drop in translation quality, making it a practical and environmentally sustainable solution for the TSPs.
- Open Access
- 11 December 2024
- Published by Taylor & Francis in Development Studies Research
- Vol. 12 (1)
Pragmatist-critical realism as a development studies research paradigm
- 11 December 2024
- Published by Taylor & Francis in Development Studies Research
- Vol. 12 (1)
Open AccessPragmatist-critical realism as a development studies research paradigm
- 1 August 2024
- Published by Springer Nature
- p. 245-257
Partners in Design: Insights from an Innovative Education for Sustainable Development Module Design Process
- 1 August 2024
- Published by Springer Nature
- p. 245-257
Partners in Design: Insights from an Innovative Education for Sustainable Development Module Design Process
- 25 May 2023
- Published by Springer Nature
- p. 1700-1705
Standardization and the Governance of Artificial Intelligence Standards
- 25 May 2023
- Published by Springer Nature
- p. 1700-1705
Standardization and the Governance of Artificial Intelligence Standards
- 10 May 2023
- Published by Association for Information Systems in Information Technology for Development
- Vol. 30 (3) , 379-399
Data collection in the global south: practical, methodological, and philosophical considerations
Data collection in resource-constrained environments, particularly in the Global South, is challenging for a variety of philosophical, theoretical, methodological, and ethical reasons. Challenges may involve using research approaches designed in the Global North that are not appropriate when using data from the social and cultural contexts in the Global South. Adopting a reflective approach, this paper examines research projects in Bangladesh, Malawi, Sierra Leone, and Rwanda and highlights the challenges encountered on these projects. Along with the problems, attempted ameliorations are discussed, and insights are offered on how the authors overcame these challenges. The paper posits that these challenges can be mitigated by adopting contextualist methodology resulting in theories that are based on local social and cultural processes. The paper proposes that a critical realist-philosophical approach and methodology are appropriate because of contextual specificity and the innate ability to alleviate problems associated with the Global North methodology and generalisable theory.
- 10 May 2023
- Published by Association for Information Systems in Information Technology for Development
- Vol. 30 (3) , 379-399
Data collection in the global south: practical, methodological, and philosophical considerations
Data collection in resource-constrained environments, particularly in the Global South, is challenging for a variety of philosophical, theoretical, methodological, and ethical reasons. Challenges may involve using research approaches designed in the Global North that are not appropriate when using data from the social and cultural contexts in the Global South. Adopting a reflective approach, this paper examines research projects in Bangladesh, Malawi, Sierra Leone, and Rwanda and highlights the challenges encountered on these projects. Along with the problems, attempted ameliorations are discussed, and insights are offered on how the authors overcame these challenges. The paper posits that these challenges can be mitigated by adopting contextualist methodology resulting in theories that are based on local social and cultural processes. The paper proposes that a critical realist-philosophical approach and methodology are appropriate because of contextual specificity and the innate ability to alleviate problems associated with the Global North methodology and generalisable theory.
- Open Access
- 7 March 2023
- Published by Taylor & Francis in Ethics & Behavior
- Vol. 34 (3) , 199-217
Grey zones and good practice: A European survey of academic integrity among undergraduate students
Good academic practice is more than the avoidance of clear-cut cheating. It also involves navigation of the gray zones between cheating and good practice. The existing literature has left students’ understanding of gray zone practices largely unexplored. To begin filling in this gap, we present results from a questionnaire study involving N = 1639 undergraduate students from seven European countries representing all major disciplines. We show that large numbers of these students are unable to identify gray area issues and lack sensitivity to the context dependence of these. We also show that a considerable proportion of students have a poor understanding of concepts like plagiarism and falsification, not only in gray zone scenarios, but also in cases of relatively clear-cut cheating. Our results are similar across the faculties and countries of study, and even for students who have attended academic integrity training. We discuss the implications of this for academic integrity training.
- 7 March 2023
- Published by Taylor & Francis in Ethics & Behavior
- Vol. 34 (3) , 199-217
Open AccessGrey zones and good practice: A European survey of academic integrity among undergraduate students
Good academic practice is more than the avoidance of clear-cut cheating. It also involves navigation of the gray zones between cheating and good practice. The existing literature has left students’ understanding of gray zone practices largely unexplored. To begin filling in this gap, we present results from a questionnaire study involving N = 1639 undergraduate students from seven European countries representing all major disciplines. We show that large numbers of these students are unable to identify gray area issues and lack sensitivity to the context dependence of these. We also show that a considerable proportion of students have a poor understanding of concepts like plagiarism and falsification, not only in gray zone scenarios, but also in cases of relatively clear-cut cheating. Our results are similar across the faculties and countries of study, and even for students who have attended academic integrity training. We discuss the implications of this for academic integrity training.
- 20 November 2022
- Published by Springer Nature
- p. 608-618
Data Collection in the Global South and Other Resource-Constrained Environments: Practical, Methodological and Ethical Challenges
- 20 November 2022
- Published by Springer Nature
- p. 608-618
Data Collection in the Global South and Other Resource-Constrained Environments: Practical, Methodological and Ethical Challenges
- Open Access
- 6 October 2022
- Published by Springer Nature in AI and Ethics
- Vol. 3 (4) , 1095-1115
Ethics and diversity in artificial intelligence policies, strategies and initiatives
A burgeoning of Artificial Intelligence (AI) technologies in recent years has led to increased discussion about its potential to address many issues considered otherwise intractable, including those highlighted by the United Nations 2030 Agenda for Sustainable Development and associated Sustainable Development Goals. In tandem with this growth in AI is an expanding body of documentation regarding how such advanced technologies should be governed and managed. Issued by a variety of sources and comprising frameworks, policies and guidelines, this body of work encompasses the legal, social, ethical and policy issues around AI. With at least 470 such documents identified, as of May 2021, in the Council of Europe’s tracker of AI initiatives, questions are emerging around the diversity of views expressed, especially regarding the influence of the Global North or Euro-American perspectives. Our previous analysis of a corpus of largely grey literature discovered blind spots regarding both gender representation and perspectives from the Global South. Expanding on that work, this paper examines a significantly extended corpus, with a focus on the role of underrepresented groups in the wider AI discourse. We find that voices from the Global South and consideration of alternative ethical approaches are largely absent from the conversation. In light of the prominence of social, cultural and ethical perspectives from the Global North, this paper explores implications for the development of standards for ethical AI. Concluding by offering approaches to incorporate more diverse ethical viewpoints and beliefs, we call for increased consideration of power structures when developing AI ethics policies and standards within these alternative socio-cultural and socio-economic contexts.
- 6 October 2022
- Published by Springer Nature in AI and Ethics
- Vol. 3 (4) , 1095-1115
Open AccessEthics and diversity in artificial intelligence policies, strategies and initiatives
A burgeoning of Artificial Intelligence (AI) technologies in recent years has led to increased discussion about its potential to address many issues considered otherwise intractable, including those highlighted by the United Nations 2030 Agenda for Sustainable Development and associated Sustainable Development Goals. In tandem with this growth in AI is an expanding body of documentation regarding how such advanced technologies should be governed and managed. Issued by a variety of sources and comprising frameworks, policies and guidelines, this body of work encompasses the legal, social, ethical and policy issues around AI. With at least 470 such documents identified, as of May 2021, in the Council of Europe’s tracker of AI initiatives, questions are emerging around the diversity of views expressed, especially regarding the influence of the Global North or Euro-American perspectives. Our previous analysis of a corpus of largely grey literature discovered blind spots regarding both gender representation and perspectives from the Global South. Expanding on that work, this paper examines a significantly extended corpus, with a focus on the role of underrepresented groups in the wider AI discourse. We find that voices from the Global South and consideration of alternative ethical approaches are largely absent from the conversation. In light of the prominence of social, cultural and ethical perspectives from the Global North, this paper explores implications for the development of standards for ethical AI. Concluding by offering approaches to incorporate more diverse ethical viewpoints and beliefs, we call for increased consideration of power structures when developing AI ethics policies and standards within these alternative socio-cultural and socio-economic contexts.