Dario Rossi
Huawei Technologies, France
TalkWhy network AI problems are hard: an industrial perspective
Google Scholar →
Co-located with ACM CoNEXT 2026 · Utrecht, The Netherlands · 7 December 2026
Network Traffic Classification – Reloaded
NTC-R 2026
Rigorous. Reproducible. Reloaded.
Network traffic classification is our flagship scenario for a broader question: How do we build trustworthy, reproducible AI for networking?
Why NTC-R?
A timely and necessary reassessment of network traffic classification.
Internet traffic classification has been a central topic in networking for more than two decades. It has evolved from traffic engineering, QoS, and application-aware management to encrypted traffic analysis, machine learning, and, more recently, AI-based approaches inspired by large language and foundation models.
Recent studies, however, have revealed major methodological weaknesses, particularly in representation learning for encrypted traffic. Reported improvements often arise from dataset leakage, shortcut learning, or unrealistic evaluation settings, raising doubts about whether models learn meaningful traffic properties or merely capture spurious correlations.
At the same time, the field still lacks widely accepted tasks, benchmarks, standardized datasets, and reproducible evaluation practices. Fragmented datasets, inconsistent pipelines, and limited reproducibility continue to slow scientific progress.
NTC-R 2026 brings together researchers from networking, machine learning, security, and systems to reassess the state of the art and help define a research agenda grounded in rigor, reproducibility, and realistic deployment assumptions.
Traffic classification is our flagship scenario, but the questions it raises - trustworthy evaluation, reproducibility, and robustness to shortcuts - cut across networking as a whole. We therefore warmly welcome contributions that carry lessons from adjacent problems in computer networks and, more broadly, from trustworthy AI for networking, wherever they sharpen how the community measures genuine progress.
Missing “ImageNet for NTC”
The research community still lacks a shared dataset and a common set of tasks on which different proposals can be fairly compared and tested. Without shared benchmarks, evaluation pipelines, and best practices, it remains difficult to distinguish genuine progress from results that exploit dataset artefacts or unrealistic assumptions.
Rather than focusing on incremental performance gains, NTC-R 2026 promotes a rigorous scientific discussion on evaluation methodology, benchmarking, reproducibility, trustworthy AI for traffic classification, and representation learning approaches grounded in networking knowledge.
Schedule
NTC-R 2026 is a half-day workshop held on the afternoon of 7 December 2026. Timing may still change slightly.
-
13:45
Opening Remarks
-
13:55Keynote
Why network AI problems are hard: an industrial perspective
-
14:45Paper
From Model-centric to Problem-centric: Revisit Machine Learning for Network Traffic Classification
-
15:00Paper
NetReplica: A Bottleneck-Regime Substrate for Reproducible Network Data Generation
-
15:15Paper
More Classes, More Problems? Scalability and Efficiency Trade-offs in Network Traffic Classification
-
15:30
Coffee Break
-
15:55Keynote
Network Traffic Classification Research to Deployment: Experiences and Use-Cases
-
16:45Paper
Rethinking Network Foundation Models: From Bigger Models to Better Data
-
17:00Paper
CoLD++: Improving Collaborative Label Denoising for Network Intrusion Detection using Label Correction
-
17:15
Closing Remarks
Accepted Papers
-
CoLD++: Improving Collaborative Label Denoising for Network Intrusion Detection using Label Correction
-
From Model-centric to Problem-centric: Revisit Machine Learning for Network Traffic Classification
-
More Classes, More Problems? Scalability and Efficiency Trade-offs in Network Traffic Classification
-
NetReplica: A Bottleneck-Regime Substrate for Reproducible Network Data Generation
-
Rethinking Network Foundation Models: From Bigger Models to Better Data
Invited Speakers
NTC-R 2026 features two invited keynotes. Abstracts will be announced closer to the event.
Vijay Sivaraman
University of New South Wales, Australia
TalkNetwork Traffic Classification Research to Deployment: Experiences and Use-Cases
Google Scholar →Topics of Interest
The workshop solicits original research papers, measurement studies, system papers, position papers, reproducibility studies, benchmark contributions, and negative results centered on Network Traffic Classification - and, more broadly, on trustworthy and reproducible AI for networking. Topics include, but are not limited to:
Benchmarking methodologies and reproducibility
Open datasets and traffic traces
Analysis of biases in existing datasets
Detection and mitigation of shortcut learning
Encrypted traffic classification
Foundation models and LLM-inspired approaches for networking
Explainable and interpretable traffic classification
Lightweight and deployable ML solutions
Adversarial robustness and evasion resistance
Measurement studies and operational experience
Reproducibility reports and artifact evaluation
Negative results and lessons learned
Shortcut-robust approaches for network traffic classification
Lessons from trustworthy AI in adjacent networking problems
Evaluation and benchmarking across networking ML tasks
Important Dates
All deadlines are Anywhere on Earth (AoE).
20 July 2026
Paper Registration Deadline
Register your paper (title and abstract).
24 July 2026
Paper Submission Deadline
Final deadline to upload the full paper.
30 August 2026
Acceptance Notification
5 October 2026
Camera-Ready Due
7 December 2026
Workshop Day
Co-located with ACM CoNEXT 2026, Utrecht, The Netherlands
Submission Instructions
NTC-R 2026 adopts a double-blind peer review process. Submissions on traffic classification are our core focus, and we equally welcome work from adjacent networking problems that advances trustworthy, reproducible AI for the field. Authors are strongly encouraged to release all supporting materials alongside their submissions.
Paper Format
- At most five (5) pages, excluding references and appendix
- Camera-ready: up to six (6) pages to address reviewers’ comments (rules for references and appendix unchanged)
- Two-column, 10pt ACM format
- Double-blind: make sure to preserve anonymity
- LaTeX users: ACM conference proceedings template
Open Science Encouraged
- Code and software artifacts
- Datasets and traffic traces
- Experimental artifacts
- Reproducibility instructions
Releasing materials is not required for acceptance, but strongly encouraged in the spirit of reproducible research.
Authors of accepted papers are expected to present their work at the workshop. Accepted papers will appear in the ACM Digital Library.
Submissions are handled via HotCRP.
Prepare your paper following the format above, then submit through the portal.
Submit on HotCRP →
Organizers
Steering Committee
Marco Mellia
Politecnico di Torino, Italy
Dario Rossi
Huawei Technologies, France
Program Chairs
Matteo Boffa
Politecnico di Torino, Italy
Francesco Bronzino
ENS de Lyon, France
Paul Schmitt
Cal Poly, USA
Technical Program Committee
Aristide Tanyi-Jong Akem
University of Southampton, UK
Giovanni Apruzzese
Reykjavik University, Iceland
Levente Csikor
A*STAR, Singapore
Nick Feamster
University of Chicago, USA
Claudio Fiandrino
IMDEA Networks Institute, Spain
Alessandro Finamore
Huawei Technologies, France
Arpit Gupta
UC Santa Barbara, USA
Prankur Gupta
Meta, USA
Minzhao Lyu
University of New South Wales, Australia
Paul Patras
University of Edinburgh, UK
Valerio Persico
University of Naples Federico II, Italy
Adrien Schoen
CNRS, France
Walter Willinger
NIKSUN, USA
Utrecht, The Netherlands
NTC-R 2026 is co-located with ACM CoNEXT 2026, held in Utrecht, The Netherlands, 7–10 December 2026: workshops on 7 December, main conference 8–10 December. NTC-R 2026 takes place on 7 December 2026.
Utrecht is easily accessible from Amsterdam Schiphol Airport (30 min by train) and sits at the heart of the Dutch rail network.
For full venue details, accommodation options, and visa information, please refer to the main conference website.
How to Register
Registration for NTC-R 2026 is handled through ACM CoNEXT 2026 and is now open. Choose either a Workshops Only or a Full Conference and Workshops registration: the Full Conference ticket alone does not include the workshops. On-site registration is not possible, so please register in advance.
Get in Touch
For questions about submissions, the program, or the workshop in general, reach out to the Program Chairs.