UCD CS academics invited to organise prestigious Dagstuhl seminars
The invitation-only (opens in a new window)Dagstuhl Seminars are highly prestigious and provide a unique format and environment for the exchange and development of ideas. Each seminar focuses on a specific area of computing research. A Dagstuhl seminar involves intense collaboration among top computer science professionals, influencing progress and direction in the science. They are attended by specialist scholars and industry experts from many countries. Seminars take place at the (opens in a new window)Schloss Dagstuhl – Leibniz Center for Informatics in Germany.
The seminars that UCD CS staff are organising are on:
- Explainable AI for Sequential Decision Making
- Machine Learning Augmented Algorithms for Combinatorial Optimization Problems
- Generative AI in Programming Education
Professor (opens in a new window)Mark Keane has been invited to co-organise and present a Dagstuhl Seminar in Germany with international collaborators from France, Netherlands and the USA. The seminar is on (opens in a new window)Explainable AI for Sequential Decision Making. It will take place at (opens in a new window)Schloss Dagstuhl – Leibniz Center for Informatics in Germany over 3 days ( 9-11 September 2024) with approximately 25 participants. Prof. Keane’s collaborators are Hendrik Baier (TU Eindhoven, NL), Sarath Sreedharan (Colorado State University – Fort Collins, USA), Silvia Tulli (Sorbonne University – Paris, FR) and Abhinav Verma (Pennsylvania State University, USA).
As we work with AI and rely on AI for more and more decisions that influence our lives, the research area of explainable AI (XAI) has rapidly developed, with goals such as increasing trust, enhancing collaboration, and enabling transparency in AI. However, to date, the focus of XAI has largely been on explaining the input-output mappings of “black box” models like neural networks, which have been seen as the central problem for the explainability of AI systems. While these models are certainly important, intelligent behaviour often extends over time and needs to be explained and understood as such. The challenge of explaining sequential decision-making (SDM), such as that of robots collaborating with humans or software agents engaged in complex ongoing tasks, has only recently gained attention. The seminar will focus on under-researched challenges that are unique to, or of particular relevance to, explainability in sequential decision-making settings. The aim is to move towards a shared understanding of the field, and to develop a common roadmap for moving it forward.
Assistant Professor (opens in a new window)Deepak Ajwani is co-organising a seminar on (opens in a new window)Machine Learning Augmented Algorithms for Combinatorial Optimization Problems, with Bistra Dilkina (USC - Los Angeles, US), Tias Guns (KU Leuven, BE), Ulrich Carsten Meyer (Goethe University - Frankfurt am Main, DE). This seminar will take place 27-31 October 2024. The seminar will bring together approximately 45 specialist scholars working at the intersection of algorithm engineering, optimisation solvers and machine learning fields. The proposal to organise this seminar was selected based on a competitive selection process.
Combinatorial optimization problems arise naturally in a multitude of crucial applications, ranging from business analytics, engineering, supply-chain optimisation, transportation, bioinformatics etc. In recent years, motivated by the success of machine learning in diverse fields, researchers have explored if learning techniques can be used to efficiently solve combinatorial optimisation problems. This is challenging because these problems have highly correlated decision variables and the correlations are long-range with very little spatial or temporal coherence. As a result, the end-to-end learning systems that take the problem instance as an input and produce the optimal solution as an output often do not generalise well to instances of larger sizes and from a different input distribution. Experts in this area have advocated for using machine learning in combination with current combinatorial optimisation algorithms to benefit from the theoretical guarantees and state-of-the-art algorithms already available. The discussion in this seminar will focus on how best to combine the machine learning techniques with algorithmic insights and optimisation solvers to solve combinatorial optimisation problem instances more scalably. These discussions are expected to accelerate the pace of research in this area and build collaborations and synergies between the researchers working in the areas of algorithm design and engineering, optimisation solvers and machine learning.
Assistant Professor (opens in a new window)Brett A. Becker is leading a seminar on (opens in a new window)Generative AI in Programming Education, with Michelle Craig (University of Toronto), Paul Denny (University of Auckland), and Natalie Kiesler (Leibniz Institute for Research and Information in Education). This seminar will take place July 27 – August 1, 2025. The seminar will bring together approximately 45 world-renowned scholars working at the intersection of computing education, human-computer interaction, software engineering, and artificial intelligence. Dr. Becker is the lead organiser of this seminar.
Generative AI stands to significantly disrupt education in general and programming education is no exception. In addition, learning to program has several unique requirements and characteristics that require specific approaches. Evidence from the past several decades on how humans learn programming supports the commonly adopted approach of having students write many small programs. Often these are checked, and feedback provided, by automated assessment tools. However, Generative AI has likely rendered this approach obsolete given that tools now exist (and are readily available and easy to use) that can solve introductory computing problems with natural language prompts (for instance a copy/paste of an exam question. Additionally it is well known that the large language models that power Generative AI tools often provide incorrect and/or biased output, and it is possible that students could become over-reliant on these tools or generate code plagiarised from online sources by the model. Such tools also often generate code that students do not understand. Educators may try to employ “AI detectors” to enforce academic integrity however recent evidence indicates that AI generated code is undetectable by standard tools for measuring code similarity.
Educators are currently taking a variety of approaches, including ignoring the issue. Generative AI is a nascent yet very rapidly developing field and new challenges and opportunities arise frequently making it extremely difficult for educators to keep pace with developments. Very recently prototype tools that leverage Generative AI to facilitate learning are appearing. However these tools have yet to be deployed or adopted at scale. New pedagogical approaches are also being developed such as, in the last few months, a new textbook which presents one approach for using Generative AI in programming education. However there is currently no evaluation of such approaches as the field is developing so rapidly.
Interest in Generative AI is growing as fast as the technology itself. A recent ACM tech talk by Porter & Zingaro had 5071 attendees registered, of which 2322 attended. A survey found that 93.8% of these attendees said that Generative AI was applicable to their work. However, given the rapid proliferation and improvement of these technologies, very little evidence exists to make definitive predictions or concrete employment of Generative AI without simply guessing how to proceed. In fact, the ACM/IEEE/AAAI CS2023 Computing Curricula includes very little on Generative AI because there isn’t enough yet known about the concrete impacts. Additionally, Generative AI is causing changes in industry. Recent reports estimate that Generative AI will add trillions to global GDP and thousands of companies are using tools powered by Generative AI such as GitHub Copilot. This will certainly change what is required of Computer Science graduates and educators need to understand what skills will be needed by their students.
This seminar aims to bring together experts and stakeholders in Generative AI, computing education, software engineering, HCI, and other computing disciplines to foster collaboration in order to chart a way forward as Generative AI continues to improve and proliferate in terms of programming capability and programming education.
Selected papers from the Seminars will be published in (opens in a new window)Dagstuhl Reports.
The 2024 seminar schedule can be found at this link (opens in a new window)https://www.dagstuhl.de/en/seminars/seminar-calendar
Published 28 March 2024 (updated 15 April)