Schedule

Welcome & Introduction

Talk: Athul Prasad

6G: Future Wireless for the AI Era

We intend to provide a holistic view from an industry perspective that includes megatrends driving technology evolution towards 6G in the AI era, new services envisioned and enabled, as well as technical requirements to realize these new services. We will present a few examples about our research related to how the emerging AI-based network optimization mechanisms enable significant performance improvements. We also discuss our vision of how the intersection of AI and mobile network infrastructure and AI infrastructure could enable new monetization capabilities for the network.

Talk: Youngjung Uh

Unlocking and Extending Prebuilt Generative AIs

Recent generative models excel at producing stunning images and videos. Unfortunately, they are trained on huge computing resources which are affordable for only big companies. How do we contribute to developing better generative models in academia without such resources? In this talk, I hope to share how we understand the pre-trained generative models and use them to achieve more precise control, higher quality. Furthermore, such understanding helps us to build architectures from them for multi-modal generation.

Talk: Yibo Yang

Recent Advances in Diffusion-Based Generative Compression

Popularized by their remarkable image generation performance, diffusion and related methods for mass transport (“diffusion” for short) have found widespread success across visual media applications. This talk will examine the recent impact of diffusion modeling on lossy data compression, where diffusion-based generative compression methods can now produce photorealistic reconstructions at extremely low bitrates. While most existing methods follow a common algorithmic framework that employs conditional diffusion for decoding, recent work has also begun to explore the use of diffusion models themselves for information transmission. Along the way we will discuss connections to inverse problem solving and rate-distortion-realism theory, examine architectural choices, and identify open research questions.

Morning Poster Session

Lunch Break

Talk: Arash Vahdat

On the Limitations of Generative Diffusion Models

Diffusion models have achieved remarkable success in generating high-quality and diverse outputs, supported by stable and scalable training procedures. However, when applied to real-world scenarios, they continue to exhibit key limitations. In this talk, I will examine several of these fundamental challenges, including slow sampling, inadequate modeling of distributional tails, and inefficiencies in training. I will also present recent advances aimed at mitigating these issues, focusing on techniques such as accelerated sampling via trajectory and distribution matching objectives, as well as improved diffusion processes designed specifically for video generation. The talk will conclude with a discussion of emerging requirements for generative models, particularly their ability to capture rare events and heavy-tailed data distributions.

Talk: Christina Chaccour

Composing Intelligence: Lessons from AI in Networks and Beyond

As AI becomes increasingly integral to cellular networks, from the RAN to the edge, we are beginning to understand what it takes to move from model development to system integration. Across deployments — from reinforcement learning for link adaptation, to early use of foundational models for anomaly detection, to modular frameworks like rApps — we have seen both the promise and the architectural friction that emerges when AI meets real-time, distributed infrastructure.

These experiences reveal more than constraints — they surface the design conditions for AI in networks: temporal dynamics, coordination across layers, system observability, and the role of domain priors. They also raise a deeper question: if foundation models like GPTs have transformed text and language understanding, what is the equivalent paradigm shift for networks? What is the representational or architectural breakthrough that will enable models to reason, adapt, and collaborate in the fabric of network systems?

We conclude with a forward-looking perspective on AI-native 6G — not as a catchphrase, but as a co-design challenge. One where models and infrastructure evolve together, and where intelligence is embedded not just at the edge, but throughout the system — distributed, agentic, and ready to act.

Afternoon Poster Session

Coffee Break

Talk: Christopher Mutschler

Taming the Radio Maze: Machine Learning for Robust and Precise Wireless Localization

The imperative need for precise positioning across urban navigation and industrial automation domains has prompted exploration into various locating methods, prominently leveraging wireless positioning systems. However, pervasive multi-path signal propagation, especially in indoor and industrial settings, complicates the estimation of the signal propagation time, the accurate estimation of which is crucial for muti-lateration. This talk focuses on the application of ML methods for wireless positioning, illustrating how channel information from measurements allows to extract valuable insights from the radio environment. This includes aspect of sim2real and transfer learning, manifold learning for self-supervised fingerprinting, and task-agnostic radio foundation models. These insights facilitate robust and precise localization of mobile objects in challenging environments. Furthermore, the discussion extends to the future landscape of 6G applications where AI could be used for a variety of new and enhanced features, e.g. ISAC sensing, enhanced localization, and further improved beam management.

Panel Session

·        Christopher Mutschler (Fraunhofer IIS)

·        Shiqiang Wang (IBM T.J. Watson)  

·        Christina Chaccour (Ericsson)

·        Juhyung Lee (Nokia)

·        Jihong Park (SUTD)

·        Deniz Gunduz (Imperial College London)

Awards & Closing