AI Symposium @ A*STAR

Prof Lishuai Li City University of Hong Kong

Abstract: Air transportation accounts for 8-12% of motorized passenger-kilometers, yet it supports 3.5% of global GDP and moves over 35% of world trade by value. Yet the existing system suffers from operational inefficiencies driven by outdated optimization models, exhibits profound vulnerability to disruptions, remains siloed from other transportation modalities, and relies on human-centric communication protocols that are incompatible with a future of autonomous flight. Emerging technologies like Advanced Air Mobility and drone logistics promise to reshape our skies, they also expose the fundamental limitations of our current infrastructure. This talk presents analytical methods designed to make air transportation systems more sustainable, resilient, integrated, and autonomous. My research focuses on developing these methods by using large-scale, real-world operational data to improve system modeling, prediction, and planning. I will explain how we combine physics-based models with machine learning, and optimization to achieve these goals. I will provide several examples, including the use of supervised and unsupervised learning to analyze data from existing airline and airport operations, identifying opportunities to improve efficiency and performance; the application of dynamic programming and reinforcement learning to plan for emerging technologies, demonstrated through our work on designing optimal route networks for urban drone delivery.

Prof Wenwu Wang University of Surrey, UK

Abstract: Large Language Models (LLMs) are increasingly being applied to audio processing, where they help interpret and generate meaningful patterns from complex sound inputs such as speech, music, and sound effects. When combined with acoustic models, LLMs offer significant potential for solving a wide range of challenges in audio processing, understanding and generation. This talk will highlight several recent developments in large audio-language models (LALMs), focusing on new algorithms and their applications to audio-centric tasks. Topics will include audio-text fusion and alignment, cross-modality audio applications, the construction of audio-language datasets, and emerging research directions in audio-language learning. We will showcase our recent work in areas such as audio generation and storytelling (e.g., AudioLDM, AudioLDM2, WavJourney), audio source separation (e.g., AudioSep), audio captioning and reasoning/question answering (e.g., ACTUAL and APT-LLMs), neural audio coding (e.g., SemantiCodec), audio editing (e.g., WavCraft), and the datasets (e.g., WavCaps, Sound-VECaps, AudioSetCaps) used to train and evaluate large audio-language models.

Biography: Wenwu Wang is a Professor in Signal Processing and Machine Learning, Associate Head of External Engagement, School of Computer Science and Electronic Engineering, University of Surrey, UK. He is also an AI Fellow at the Surrey Institute for People Centred Artificial Intelligence. His current research interests include signal processing, machine learning and perception, artificial intelligence, machine audition (listening), and statistical anomaly detection. He has (co)-authored over 300 papers in these areas. His work has been recognized with more than 15 accolades, including the Audio Engineering Society Best Technical Paper Award (2025), IEEE Signal Processing Society Young Author Best Paper Award (2022), ICAUS Best Paper Award (2021), DCASE Judge’s Award (2020, 2023, and 2024), DCASE Reproducible System Award (2019 and 2020), and LVA/ICA Best Student Paper Award (2018). He is a Senior Area Editor (2025-2027) of IEEE Open Journal of Signal Processing and an Associate Editor (2024-2026) for IEEE Transactions on Multimedia. He was a Senior Area Editor (2019-2023) and Associate Editor (2014-2018) for IEEE Transactions on Signal Processing, and an Associate Editor (2020-2025) for IEEE/ACM Transactions on Audio Speech and Language Processing. He is Chair (2025-2027) of the EURASIP Technical Area Committee on Acoustic Speech and Music Signal Processing, and an elected Member (2021-2026) of the IEEE SPS Signal Processing Theory and Methods Technical Committee. He was the elected Chair (2023-2024) of IEEE Signal Processing Society (SPS) Machine Learning for Signal Processing Technical Committee, and a Board Member (2023-2024) of IEEE SPS Technical Directions Board. He has been on the organising committee of INTERSPEECH 2022, IEEE ICASSP 2019 & 2024, IEEE MLSP 2013 & 2024, and SSP 2009. He was Technical Program Co-Chair of IEEE MLSP 2025. He has been elected to IEEE Fellow for contributions to audio classification, generation and source separation. He has been an invited Keynote or Plenary Speaker on more than 20 international conferences and workshops.

Assoc Prof Dongjin Song University of Connecticut

Abstract: Recent progress in time series analysis has been driven by deep learning and, more recently, by foundation models and large language models. However, most existing systems still behave as pattern-matching systems rather than knowledge-informed reasoners. In this talk, I will present a line of work toward knowledge- and context-aware time series models that integrate retrieval, language-based reasoning, and multi-modal views. I will first introduce TS-RAG, a retrieval-augmented framework that equips time series foundation models with a dedicated temporal knowledge base and an Adaptive Retrieval Mixer, enabling zero-shot forecasting that adapts to distribution shifts while remaining interpretable through retrieved exemplar trajectories. Next, I will present TimeXL, which combines a prototype-based multimodal encoder with three LLM agents for prediction, reflection, and refinement, forming a closed loop that jointly improves accuracy and human-centric explanations in domains such as finance, healthcare, and weather. Finally, I will discuss DMMV, a decomposition-based multimodal view framework that fuses numerical and visual (imaged) representations of time series via large vision models, carefully leveraging their inductive biases to improve long-term forecasting across diverse benchmarks. Together, these results point toward a new generation of knowledge-informed time series models that integrate knowledge, modality, and reasoning to achieve transparent, trustworthy temporal intelligence in high-stakes applications.

Biography: Dongjin Song is an Associate Professor in the School of Computing at the University of Connecticut. He was previously a Research Stag Member at NEC Labs America in Princeton, NJ, from July 2016 to July 2020. He received his Ph.D. degree in the ECE Department from the University of California, San Diego (UCSD) in 2016. His research interests include machine learning, data science, deep learning, and related applications for time series analysis and graph representation learning. His work has been published in premier conferences in artificial intelligence and data science, including NeurIPS, ICML, ICLR, KDD, ICDM, SDM, AAAI, IJCAI, CVPR, and ICCV. Four of his papers, i.e., DARNN (IJCAI 2017), HetGNN (KDD 2019), MSCRED (AAAI 2019), and Empowering TS with LLMs(IJCAI 2024) have been recognized as the most influential papers by PaperDigest.org. Dr. Song is the recipient of several prestigious awards, including the NSF CAREER Award (2024), the Frontiers of Science Award in Computer Science (2024), the UConn AAUP Excellence Award for Research and Creativity – Early Career (2025), the AI 2000 Most Influential Scholar Award Honorable Mention - Data Mining (2025), the NEC Faculty Research Award (2025), and Best Paper Award (3rd Place CCC Award) at BlueSky Track of ICDM 2025. He serves as an Associate Editor for Neural Networks and Pattern Recognition, and as an area Chair/Senior PC for top conferences such as NeurIPS, ICML, ICLR, AAAI, IJCAI, ICDM, CIKM and PAKDD.

Asst Prof Ziwei Wang Nanyang Technological University

Abstract: Vision-language-action models achieve high generalization ability and success rate due to the huge model parameters and large-scale training data. However, many applications such as manufacturing, household service and warehouse management require mobile manipulation where robots should interact with objects in different locations. Deploying VLA models in mobile manipulation is still limited as current VLA models are designed for fixed-base manipulation. To empower VLA models for mobile manipulation, we efficiently adapt the VLA models by 1) designing whole-body motion planning framework to achieve desirable manipulation trajectories from VLA, and (2) building geometric scene graph representation for base docking point selection. Our robotic manipulation system significantly broadens the application scenarios of VLA models where robots are required to complete tasks with mobility.

Biography: Ziwei Wang is currently an Assistant Professor in School of Electrical and Electronic Engineering, Nanyang Technological University (NTU). Before joining NTU, he was a postdoc fellow in Robotics Institute, Carnegie Mellon University. He received his Ph.D and the B.S degrees from the Department of Automation, Tsinghua University in 2023 and the Department of Physics, Tsinghua University in 2018 respectively. His research goal is to design foundation models (FMs) for robotics including grounding FMs to the physical scene and deploying FMs in resource-limited robots. He has published over 50 scientific papers in top-tier conferences and journals of AI, robotics and computer vision. He serves as a regular reviewer member for a variety of conferences and journals.