Keynote Speaker I

Gerhard Heyer
Leipzig University and
Corresponding member of the Saxonian Acamedy of Sciences.

Context volatility and the negotiation of meaning

The meaning of words is not static but depends on contexts of use, and changes over time. But why do the contexts change in which a word is used?
In most cases, it is possible to identify periods of significant contextual shifts during which clues to the possible causes can be found. Examples of such periods include: controversial debates in which the proponents and opponents of a position use key words in completely different contexts; specific events such as natural disasters, historical events, or technological changes, in which not only are new words introduced, but old words are also used in new contexts; cyclical events such as the “Olympics,” in which the location of the event and the names of the participants or affected actors change at intervals determined by the event.
The talk will introduce the notion of 'context volatility' as a measure of the rate of contextual change for words over a given period of time. This concept can indicate periods of controversial debate or events that attract public attention. Examples will be provided.
It will then present a game-theoretical approach to model the negotiation of meaning to explain the shifts of meaning over time. 

Gerhard Heyer is Professor Emeritus of Natural Language Processing at the Computer Science Department of Leipzig University and a corresponding member of the Saxonian Academy of Sciences. Before working at the university, he worked as a systems specialist and head of research and development in industry and at a company of his own. His research focuses on research data infrastructures, automatic semantic processing of text, and applications of text mining in the digital humanities, among others. His book on text mining (new edition 2022, https://link.springer.com/book/9783658359683) is the German standard textbook on this current topic.

Keynote Speaker II

WooKey Lee,
Department of Industrial Engineering at Inha University, South Korea

WooKey Leeis a Professor in the Department of Industrial Engineering at Inha University, South Korea, and a leading researcher in the convergence of Artificial Intelligence and Database Systems. He has served as Director of the VOICE AI Institute since 2019 and as Director of an AI Research Institute established by the Korean government since 2022.
He earned his Ph.D. in Distributed Database Systems from Seoul National University and completed his MSE at Carnegie Mellon University. Professor Lee has held leadership positions in the Korea Database Society and contributed to numerous scientific journals as an Editor-in-Chief and editor. He also serves as Vice Chair of the IEEE BigComp Steering Committee and a member of the IEEE TCDE Executive Committee.
His research interests include advanced AI applications, signal processing, data architecture, deep learning for language processing, patent classification, and face generation. His work has been published in leading venues, including CVPR, ACL, ICDM, and Nature Communications. He has received more than 30 Best Paper Awards and holds over 100 patents.
Professor Lee's extensive research and leadership contributions make him a distinguished speaker for AUTSYS 2026.

Physical AI with VLA Technology

Physical AI represents artificial intelligence that seamlessly perceives, reasons, and acts within the real world, with Vision-Language-Action (VLA) models rapidly emerging as its foundational paradigm. Inheriting rich semantic representations from vision-language models, VLAs carry the distinct advantage of executing natural-language instructions while generalizing across diverse tasks and physical platforms. However, current state-of-the-art architectures face a critical bottleneck: despite accurately interpreting high-level commands, they often fail during physical execution.
This presentation addresses contact blindness, which refers to the fundamental limitation of RGB-centric policies in discerning when, where, and whether physical contact takes place. To overcome this challenge, vision-based contact perception is introduced as a vital primitive for physically grounded execution. As a concrete instantiation, I will present the CVPR 2026 research on real-time multimodal fingertip contact detection through depth and motion fusion, demonstrating that physical contact can be visually inferred without relying on specialized tactile sensors or instrumented surfaces. Furthermore, four concrete strategies for embedding lightweight contact awareness into VLA models will be outlined: (1) contact-event tokens for policy gating, (2) auxiliary contact supervision during post-training, (3) a fast verification layer within hierarchical control, and (4) automated contact-labeling engines for learning from human video.
Ultimately, these insights outline a scalable, hardware-light roadmap for building VLA systems that bridge high-level understanding with verifiable physical success.