Academic Salon on Networked Systems and AI
Organisation by Technical University of Munich
Dates
Tuesday Afternoon 31 March and Wednesday Morning 01 April 2026
Location
TUM-IAS Institute of Advanced Studies Lichtenbergstraße 2a 85748 Garching near Munich Germany
Aims and Scope
The rapid evolution of large-scale AI systems is transforming how we design, operate, and secure modern digital infrastructures. As AI models, data pipelines, and intelligent agents become deeply networked, new architectures are required to support demanding workloads, distributed intelligence, and privacy-preserving computation across heterogeneous environments. This event brings together leading European researchers working at the intersection of networking, systems, and AI, with a focus on three converging movements: (1) network architectures designed for AI and agentic protocols, (2) distributed and trustworthy AI executed across networked systems federated platforms, and (3) data-driven AI frameworks that enable intelligent networks, large-scale measurement, and high-performance orchestration. (4) A final thread addresses the pressing need to build energy-efficient and sustainable AI data centers, ensuring that emerging AI capabilities remain environmentally viable. Through four themed sessions, the event aims to provide a comprehensive and forward-looking perspective on the technical, scientific, and architectural foundations of next-generation networked AI ecosystems.
Sessions
Sessions are planned consisting of regular presentations followed by Q&A, and panel discussions. Possible sessions that address the above outlined aims of the academic salon could be organized along the following content grouping.
Day 1 — Afternoon
Session 1 — Networked AI Architectures and Network for AI
This session investigates how modern networks must evolve to support demanding AI workloads, focusing on agentic protocols, secure and accelerated datapaths, and high-throughput transport mechanisms. It highlights architectures and prototypes that improve inference performance, protect agent interactions, and leverage eBPF, SmartNICs, and RDMA for scalable multi-agent AI systems.
Possible presentations on the following topics:
- Acceleration for AI Agents
- eBPF programmability for networked AI
- Panel discussion
Session 2 — Distributed AI and Agentic Protocols
This session addresses the principles and mechanisms required to execute AI securely and reliably across multiple sites, organizations, and trust domains. It covers agentic protocols such as A2A and MCP for interaction of LLM agents, federated inference, decentralized learning, privacy-preserving verification of neural networks, and confidential computing architectures for attestable and private machine learning.
Possible presentations on the following topics:
- Federated Inference and Agentic Protocols
- Industrial Edge Networks with Distributed AI
- Decentralized Learning, In-Network Machine Learning
- Panel discussion
Day 2 — Morning
Session 3 — AI for Networks
This session explores how AI can be applied to networks themselves, enabling large-scale measurement, intelligent orchestration, reinforcement learning for routing and path selection, and digital twins for automated network operations. It emphasizes transforming high-volume telemetry into actionable intelligence.
Possible presentations on the following topics:
- Digital Twins for High-Performance Network Orchestration
- AI-based Network Modelling and Optimization
- Network Data Generation for training domain-specific models
- Panel discussion
Session 4 — Energy-Efficiency, Sustainability / Panel Session
This session brings together researchers and industry practitioners to discuss strategies for reducing the energy footprint of AI infrastructures. Short motivating talks will highlight advances in energy-aware inference, datacenter-level optimization, and reinforcement-learning-based resource management. The panel will address sustainability challenges, operational trade-offs, and opportunities for greener large-scale AI.
Possible presentations and panel discussions on the following topics:
- Energy Efficiency in AI Inference
- Decentralized AI and Energy-Optimisation
- Panel Discussion - Possible topic: Approaches to Energy-Optimised AI Deployment
Persons proposed to be invited
Persons proposed to be invited for a presentation include
- Gustavo Alonso (ETH Zurich, CH) - GPU processing, Hardware acceleration, SmartNICs
- Marios Avgeris (University of Amsterdam, NL) - In-Network Machine Learning
- Pere Barlet-Ros (UPC Barcelona, ES) - Graph Neural Networks and AI
- Paolo Bellavista (University of Bologna, IT) - Edge-Cloud Networks and AI
- Philippe Buschmann (Siemens, DE) - Distributed AI interference in industrial scenarios
- Chunyang Chen (TUM) - Knowledge transfer across LLMs of different scales
- György Dán (KTH, SE) - AI and Network Security
- Fabien Geyer (Airbus, DE) - AI and Performance Prediction
- Hamed Haddadi (Imperial College London, UK): Distributed AI
- Shashikant Ilager (University of Amsterdam, NL) - Energy-Efficient AI
- Anne-Marie Kermarrec, Rishi Sharma (EPFL, CH) - Decentralized AI
- Marios Kogias (Imperial College London, GB) - eBPF and AI applications
- Michael Menth (University of Tübingen, DE) - eBPF for Service Function Chaining
- Johannes Späth (TUM) - Data Generation for AI
- Kurt Tutschku (Blekinge Inst. of Technology, SE) - Digital Twins and Multi-User Human-Centered Intelligent Realities
