Gerald Friedland 
Amazon, USA

Data as a Sufficient but Not Necessary Condition for Machine Learning

​

Abstract:

For decades we believed that algorithms were the key to prediction. From the Perceptron to Support Vector Machines, from Gaussian Mixture Models to Random Forests, the competition was about algorithms — until Banko and Brill (2001) demonstrated empirically that "there is no data like more data": with enough data, consistent learners converge toward the same Bayes-optimal limit, so the choice of algorithm matters less and less. The hunt for data began, and it was believed that "whoever owns the data, owns the market." Now, however, foundation models show us that — at least in the language, speech, and computer vision domains — data is no longer even the most valuable asset. Drawing on my experiments performed for Mitra, Amazon's frontier tabular foundation model, this talk illuminates why data is only the tip of the iceberg — even for tabular data — and what future research directions this implies for the field of computer science.

 

Bio:

​Dr. Gerald Friedland is a Principal Scientist at AWS and adjunct professor at the University of California, Berkeley.

 

Dr. Friedland has published about 200 scientific publications, including two books and is a part-time Lecturer at the EECS department of the University of California. Berkeley. He is the recipient of several research and industry recognitions, among them the European Academic Software Award and the Multimedia Entrepreneur Award by the German Federal Department of Economics. He also led the team that won the ACM Multimedia Grand Challenge in 2009. Dr. Friedland received his doctorate (summa cum laude) and master’s degree in computer science from Freie Universitaet Berlin, Germany, in 2002 and 2006, respectively.

​

Gerald is also an avid software and hardware developer and is known for open-source contributions, such as SIOX (Simple Interactive Object Extraction), which has become the open-source standard algorithm for interactive image cut and paste used in applications like GIMP or Inkscape.

​

​

Max Mühlhäuser 
Technical University of Darmstadt, Germany

Agentic Multimodal AI for Urban Planning and Future Cities

​

Abstract:

Cities worldwide face new challenges at a fast pace due to growing or aging populations, economic transformations (e.g., home delivery replacing shopping & dining) and geo-political tensions (increas-ing disruption risks). In reply to these challenges, *resilience* has become the new Leitmotif of urban planning & development and *Urban XR* holds the promise to “redefine the downtown experience”.

 

W.r.t. urban planning, agentic AI and language models will radically reshape the daily work of urban planners/decision-makers. As part of this development, Urban Digital Twins (UDTs) have long been advocated as a key to smarter future cities and better & cheaper urban solutions.

​

The keynote will resume pressing urban challenges and introduce Agentic AI and UDTs for urban planning/decision-making. A key problem of preset UDTs is their focus on 3D models as data anchors: Urban planning is equally concerned with urban structures and urban dynamics, calling for 4-dimen-sional UDTs in which the time dimension lives alongside the 3D structures. As to Urban Agentic AI, its potential for evidence-based urban planning & management remains underexploited. The keynote will present an architecture that integrates 4D UDTs with Agentic Multimodal AI for evidence-based urban planning and management. The concept and architecture will be demonstrated through a use case implemented by the speaker and his team.

​

The keynote will conclude with an outlook on future cities, showing that Agentic Multimodal AI with 4D-UDTs can be furthered into a general concept for all citizens and urban stakeholders in daily life: this path will lead to interactive augmented urban realities as the new normal for future cities.

 

Bio:

Max Mühlhäuser is head of the Telecooperation Lab at the Technical University of Darm-stadt. After his retirement as a full professor, he continues to work as a contracted professor emeri-tus at his university and abroad. Max is an academian with the German National Academy of Science and Engineering, IEEE Fellow and ACM distinguished member. His lab conducts research on smart spaces with a focus on smart cities and infrastructures. He made extensive contributions to these fields through innovations in Proactive AI, Human-Computer Interaction, Distributed Systems, and PST (Privacy, Cyber Security, and Trust). He held or holds key positions in several large collaborative research centers, e.g., as directorate member of the Center for Resilient Digital Cities (emergenCITY) and as spokesperson for the Doctoral School on Privacy and Trust. Max founded and managed indus-trial research centers and worked as either a professor or a visiting professor at universities in Eu-rope, North America, and Australia. He published over 800 peer-reviewed articles and was or is ac-tive as chair/member of numerous scientific conference program and organization committees, as editorial board or SC chair/member, reviewer, or guest editor for renowned scientific journals like ACM IMWUT, ACM ToIT, IEEE T-MM, ACM Multimedia, and Elsevier PMC.

​

 

​

​

​