AI, ML and Friends is a weekly seminar series within the School of Computing on Artificial Intelligence, Machine Learning, and related topics. We are open to attendees and presenters external to the school. Please sign up to the mailing list to receive weekly announcements including zoom details, and email the seminar organiser to schedule a talk.

Upcoming Seminars

13 July 2026, 12:00#

Exploring Machine Learning Techniques for Automatic Story Generation#

Speaker: Yuan Ma

Abstract: Recent advances in large language models (LLMs) have significantly improved the performance of Natural Language Processing (NLP) systems across a wide range of language generation tasks. Despite these achievements, generating coherent, creative, and meaningful long-form narratives remains a major challenge for artificial intelligence. Story generation requires not only linguistic fluency but also narrative planning, character consistency, contextual understanding, and the ability to convey human values and emotions. This research investigates automatic story generation for exploring machine creativity and improving the narrative capabilities of LLMs.

Bio: Yuan Ma is a first-year PhD candidate at the Australian National University (ANU) supervised by Hanna Suominen and Patrik Haslum. His research focuses on natural language processing and machine learning for automatic story generation.

Where: Building 145, room 3.41

20 July 2026, 11:00#

Sign Language: Towards Sign Understanding for Robot Autonomy#

Speaker: Dr. Nicky Zimmerman

Abstract: Navigational signs are common aids for human wayfinding and scene understanding but are underutilized by robots. We argue that they benefit robot navigation and scene understanding, by directly encoding privileged information on actions, spatial regions, and relations. Interpreting signs in open-world settings remains a challenge owing to the complexity of scenes and signs, but recent advances in vision-language models (VLMs) make this feasible. To advance progress in this area, we introduce the task of visual sign grounding, which maps semantic instructions on signs to corresponding scene elements and navigational actions. We also outline different applications, such as localization and navigation, which benefit from the spatial-symbolic information encoded by navigational signs.

Bio: Nicky is a Postdoctoral Research Fellow in National University of Singapore working with Prof. David Hsu. Her research focuses on open world navigation. She graduated as a PhD in Computer Science from Università della Svizzera italiana in 2024, where she also completed her MSc. Previously, she worked for General Motors and Intel.

Where: Brian Anderson Building, A105

21 July 2026, 11:00#

Commonsense Robot Planning with LLMs and Tree Search#

Speaker: Prof. Panpan Cai

Abstract: Robots operating in human environments must plan under uncertainty while reasoning with commonsense knowledge that is often implicit, incomplete, or only expressed through natural language. Recent Large Language Models (LLMs) provide powerful semantic priors for robot planning, but their open-ended reasoning is difficult to integrate with structured decision making, long-horizon planning, and uncertainty-aware execution. In this talk, I will present a series of works that bridge LLMs with symbolic, tree-search-based planning for commonsense robotic task planning. First, I will introduce Tru-POMDP, which combines LLM-generated structured hypotheses with principled POMDP planning, enabling robots to reason over ambiguous instructions, hidden object locations, and open-world uncertainty through belief-space tree search. Next, I will present UniDomain, a framework that learns reusable symbolic planning domains from large-scale robot demonstrations, allowing robots to generalize to unseen manipulation tasks by composing domain knowledge instead of relying on handcrafted planning models. Finally, I will introduce PO-PDDL, a symbolic representation and learning framework that extends PDDL to model partial observability and stochastic action outcomes, enabling reusable belief-space planning models to be learned directly from robot execution videos. Overall, the talk argues that the future of robot planning lies not in replacing classical planning with LLMs, but in tightly integrating foundation models, symbolic world models, and principled search algorithms. Such an integration provides a practical path toward robots that can reason with commonsense, adapt to uncertainty, and execute complex long-horizon tasks in open-world environments.

Bio: Panpan Cai is an Associate Professor at the School of Artificial Intelligence, Shanghai Jiao Tong University, and a Full-time Faculty Member at the Shanghai Innovation Institute. Prior to joining SJTU, she was a postdoctoral researcher in the Department of Computer Science at the National University of Singapore (NUS), and received her Ph.D. from Nanyang Technological University (NTU), Singapore. Her research interests lie in robot planning, robot learning, and their integration. She develops algorithms that integrate foundation models, symbolic reasoning, and probabilistic planning to enable robots to perform long-horizon tasks in open-world environments, such as urban city roads and homes. She has published in leading robotics and AI venues, including The International Journal of Robotics Research (IJRR), IEEE Transactions on Robotics (T-RO), Robotics: Science and Systems (RSS), Conference on Robot Learning (CoRL), NeurIPS, ICRA, and IROS. Her work was a finalist for the CoRL Best Paper Award in 2022. She currently serves as an Associate Editor of IEEE Transactions on Robotics (T-RO) and has served as an Area Chair for RSS 2026 and an Associate Editor for ICRA 2024.

Where: Brian Anderson Building, A105

03 September 2026, 11:00#

Preaching to the converted. Tailoring large language model dialogue to differentiate idea endorsement and system evaluation.#

Speaker: Prof. Giles Hirst

Abstract: Large language models (LLMs) are not only a relevant persuasive tool for politics, conspiracy theories, and contested social issues, but also for ‘big business’. We examine how LLMs monetize and persuade in interactions, impacting (a) perceptions and support of new ideas (message) and (b) attitudes toward the Artificial Intelligence (AI) system (messenger). We theorize and support a two-sided commercial transaction: an LLM “speaking to” a user’s moral values lead users to evaluate both message and messenger more positively: they are more likely to endorse, use, and pay for a novel concept and they perceive the LLM understands them better. Morally congruent framing sells the object as well as enhances perceptions of the user-AI relationship. Crucially, for message recipients, this effect is most pronounced for those with more polarized political identities, addressing who is most malleable through AI influence.

Bio: Giles Hirst is Professor of Leadership at the Research School of Management, The Australian National University, and Fellow at the Judge Business School, University of Cambridge. An organisational psychologist by training, he completed his PhD at the Melbourne Business School. Developing creative ideas is one of mankind’s greatest gifts, and this is one of Giles main interests helping individuals and leaders unlock their creativity in partnership with technology. Giles’ research examines leadership, creativity, and innovation—particularly how leaders and employees can harness technology to enhance creativity and inclusion at work. His work also explores social impact themes such as improving refugee employment outcomes and addressing precarious work. He has published widely in top management journals including the Academy of Management Journal, Academy of Management Review, Leadership Quarterly, Journal of Management, and Journal of Applied Psychology. Recognised among the world’s top 2% most-cited scholars, Giles serves as Associate Editor of the Journal of Organizational Behavior and Consulting Editor for the Journal of Applied Psychology. An award-winning educator, Giles has received Vice-Chancellor’s Awards for Teaching Excellence and led leadership programs with government and commerce. He brings rich industry experience, having worked as a management consultant, served on not-for-profit social housing boards, and commercialised new business ventures. His programs help leaders connect with purpose, use their strengths, and achieve greater impact.

Where: Building 145, room 1.33

24 September 2026, 11:00#

Informed Machine Learning and Explainability for Binary Image Processing via Mathematical Morphology#

Speaker: Dr. Diego Marcondes

Abstract: The explainability of machine learning methods has recently become an issue in virtually all domains of application, and important research lines aiming to explain high-performance black boxes have been explored within the explainable AI initiative. Alternatively, this talk presents a different line of research focused on developing fully interpretable models that can evolve into high-performance methods. In particular, we will discuss how informed machine learning, characterised by the insertion of strong domain knowledge into the model design, can be leveraged to obtain interpretable methods. The main concepts will be discussed in the context of binary image processing and recent developments in mathematical morphology, in particular the discrete morphological neural networks.

Bio: Diego Marcondes earned a BS in Statistics and a PhD in Applied Mathematics from the University of São Paulo, Brazil. He is currently a research fellow at the Mathematical Data Science Centre, Mathematical Sciences Institute, The Australian National University and member of the IRL FAMSI. Previously, he was a postdoc at the Computer Science Department, Institute of Mathematics and Statistics, University of São Paulo (2022-2024), and a visiting postdoctoral scholar at the Department of Electrical and Computer Engineering, Texas A\&M University (2023-2024). His research interests in data science are at the intersection of probability theory, statistics, applied mathematics and computer science. In particular, he is interested in developing learning methods with a strong mathematical basis which do not only have a high performance, but are controllable and interpretable, with applications to scientific problems, and signal and image processing.