The SNAI 2025 proceedings contain 38 papers accepted from 143 submissions (26.6% acceptance rate). Papers were distributed across the six tracks as follows: Track 1 (8), Track 2 (7), Track 3 (6), Track 4 (6), Track 5 (6), Track 6 (5). The full proceedings are archived in the LiteConf open-access repository at liteconf.org/cfp/snai-2025/.

Award-winning papers are marked with a gold badge.

Track 1 — AI-Driven Network Management and Automation (8 papers)

• Learned Congestion Control in the Wild: A Five-Year Retrospective Best Paper Hyunwoo Ahn, Sara Bertrand, Wei Chen — KAIST; Sorbonne Université; Peking University
• Intent-Based Routing with Retrieval-Augmented Policies Mei Han, Liang Liu — Zhejiang University; Tsinghua University
• Closing the Loop on Optical Network Reconfiguration with Offline Reinforcement Learning Karim Hassan, Jonas Fischer — ETH Zürich; TU München
• Self-Driving Backbone Networks: Lessons from a Year of Closed-Loop Deployment Ricardo Almeida, Pavlos Vassileiou — Universidade de Lisboa; FORTH
• Telemetry Summarization for Root-Cause Analysis at Carrier Scale Hiroshi Iwasaki, Yui Tanaka — NTT; Osaka University
• Graph-Aware Traffic Engineering with Multi-Agent Reinforcement Learning Lukas Steiner, Anders Dahl — TU Berlin; Aalborg University
• Configuration Synthesis for SD-WAN: A Constraint-Learning Approach Rohit Bhardwaj, Ravi Ramanathan — IIT Delhi; IIT Madras
• Anomaly-Aware Scheduling for 5G Uplink Fatima Al-Saleh — KACST

Track 2 — Edge and In-Network Intelligence (7 papers)

• Partitioning LLM Inference Across the Edge and the Switch Fabric Karim Hassan, Helena Pereira — ETH Zürich; INESC-ID
• Line-Rate Feature Extraction with P4 and Quantized Trees Adriana Kowalski, Lukas Steiner — University of Warsaw; TU Berlin
• Latency-Aware Serving Across Heterogeneous Edge Devices Sara Bertrand, Ana Novak — Sorbonne Université; University of Ljubljana
• SmartNIC Offload for Distributed Transformer Inference Wei Chen, Liang Liu — Peking University; Tsinghua University
• In-Network Aggregation for Federated Edge Training Mei Han, Yui Tanaka — Zhejiang University; Osaka University
• Co-Designing Pipelines and Models for Programmable NICs Pavel Orlov, Jonas Fischer — Skoltech; TU München
• Adaptive Quantization for Edge Inference Under Memory Pressure Hyunwoo Ahn — KAIST

Track 3 — Distributed and Federated Learning over Networks (6 papers)

• Gradient Compression under Bandwidth-Varying WAN Links Ravi Ramanathan, Rohit Bhardwaj — IIT Madras; IIT Delhi
• Straggler Mitigation through Topology-Aware Scheduling Sara Bertrand, Carolina Meneses — Sorbonne Université; Universidad de los Andes
• Decentralized Training over Peer-to-Peer Topologies Ana Novak, Pavel Orlov — University of Ljubljana; Skoltech
• Privacy-Preserving Aggregation under Heterogeneous Client Bandwidth Helena Pereira, Ricardo Almeida — INESC-ID; Universidade de Lisboa
• Split Learning Across Edge and Cloud: An Empirical Study Mei Han, Liang Liu — Zhejiang University; Tsinghua University
• Communication-Efficient Federated Fine-Tuning of Foundation Models Yusuf Mwangi, Karim Hassan — University of Cape Town; ETH Zürich

Track 4 — Machine Learning for Network Security (6 papers)

• Encrypted-Traffic Classification Under Distribution Shift Best Student Paper Adriana Kowalski, Jonas Fischer — University of Warsaw; TU München
• Adversarial Robustness of Learning-Based Intrusion Detectors Hiroshi Iwasaki, Pavlos Vassileiou — NTT; FORTH
• Measuring the Longitudinal Drift of DNS Abuse Classifiers Lukas Steiner, Anders Dahl — TU Berlin; Aalborg University
• Cross-Protocol Attack Attribution with Graph Neural Networks Fatima Al-Saleh — KACST
• Federated Threat Hunting across Organizational Boundaries Carolina Meneses, Helena Pereira — Universidad de los Andes; INESC-ID
• Measurement and Detection of BGP Route Manipulation Campaigns Wei Chen, Hyunwoo Ahn — Peking University; KAIST

Track 5 — Programmable Networks and Data-Plane ML (6 papers)

• A Compiler Framework for Deploying Quantized Models onto P4 Targets Lukas Steiner, Adriana Kowalski — TU Berlin; University of Warsaw
• Hardware/Software Co-Design for Tree Inference on Programmable NICs Jonas Fischer, Pavel Orlov — TU München; Skoltech
• eBPF Pipelines for Real-Time Traffic Classification Anders Dahl, Lukas Steiner — Aalborg University; TU Berlin
• Feature Engineering at Line Rate with Match-Action Primitives Hyunwoo Ahn, Adriana Kowalski — KAIST; University of Warsaw
• Benchmarking Data-Plane ML: Workloads, Metrics, and Pitfalls Pavel Orlov, Karim Hassan — Skoltech; ETH Zürich
• Stateful Data-Plane Telemetry for ML Training Pipelines Ana Novak, Jonas Fischer — University of Ljubljana; TU München

Track 6 — Generative AI and Foundation Models for Networking (5 papers)

• Retrieval-Augmented Generation for Network Incident Response Best Artifact Ricardo Almeida, Helena Pereira — Universidade de Lisboa; INESC-ID
• Closed-Loop Network Remediation with Tool-Using LLM Agents Pavlos Vassileiou, Ricardo Almeida — FORTH; Universidade de Lisboa
• Foundation Models for Encrypted Traffic Understanding Hyunwoo Ahn, Fatima Al-Saleh — KAIST; KACST
• Evaluating LLM Agents for Configuration Synthesis: A Benchmark Wei Chen, Mei Han — Peking University; Zhejiang University
• Prompt Engineering for Operator-Facing Network Copilots Liang Liu, Pavlos Vassileiou — Tsinghua University; FORTH

Awards

• Best Paper — "Learned Congestion Control in the Wild: A Five-Year Retrospective" (Track 1).
• Best Student Paper — "Encrypted-Traffic Classification Under Distribution Shift" (Track 4).
• Best Artifact — "Retrieval-Augmented Generation for Network Incident Response" (Track 6). Inaugural award in 2025.
• Distinguished Reviewer — awarded to one member of the program committee for outstanding review contributions.