ACM SIGCOMM 2019, Beijing, China

ACM SIGCOMM 2019 Workshop on Network Meets AI & ML (NetAI 2019)

Workshop Program

Click here to access the proceedings

  • Friday, August 23, 2019, Shangri-La Hotel

  • 09:00am - 10:00am Opening + Keynote

  • Keynote: How to Make Decisions (Optimally)

    Siddhartha Sen (Microsoft Research)

    • Abstract: I will describe an agenda for applying artificial intelligence to networked systems that is minimally disruptive, synergistic with human solutions, and safe. First, I will develop a paradigm that combines reinforcement learning with the ability to ask counterfactual ("what if") questions about any decision-making system, provided there is sufficient randomness in the decisions. We will apply this methodology to infrastructure systems in Azure and face some interesting challenges and opportunities. Then, I will propose an abstraction called a “safeguard” that protects an AI system from violating a safety specification, while allowing the system (and the safeguard) to adapt and learn.


      Bio: Siddhartha Sen is a Principal Researcher in the Microsoft Research New York City lab, and previously a researcher in the MSR Silicon Valley lab. He uses data structures, algorithms, and machine learning to build more powerful distributed systems. His current mission is to optimize cloud infrastructure decisions in a way that is minimally disruptive, synergistic with human solutions, and safe. Siddhartha received his BS degrees in computer science and mathematics and his MEng degree in computer science from MIT. From 2004-2007 he worked as a developer at Microsoft and built a network load balancer for Windows Server. He returned to academia and completed his PhD from Princeton University in 2013. Siddhartha received the inaugural Google Fellowship in Fault-Tolerant Computing in 2009, the best student paper award at PODC 2012, and the best paper award at ASPLOS 2017.


  • 10:00am - 10:30am Coffee break

  • 10:30am - 12:00pm Session 1: Optimization & Formalization

    Session chair: Junchen Jiang (Tsinghua Univ, China)

  • DeePCCI: Deep Learning-based Passive Congestion Control Identification

    Constantin Sander (RWTH Aachen University), Jan Rüth (RWTH Aachen University), Oliver Hohlfeld (Brandenburg University of Technology), Klaus Wehrle (RWTH Aachen University)

  • Contextual Multi-Armed Bandits for Link Adaptation in Cellular Networks

    Vidit Saxena (KTH Royal Institute of Technology), Joakim Jalden (KTH Royal Institute of Technology), Joseph E. Gonzalez (UC Berkeley), Mats Bengtsson (KTH Royal Institute of Technology), Hugo Tullberg (Ericsson AB), Ion Stoica (UC Berkeley)

  • RL-Cache: Learning-Based Cache Admission for Content Delivery

    Vadim Kirilin (IMDEA Networks Institute), Aditya Sundarrajan (UMass, Amherst), Sergey Gorinsky (IMDEA Networks Institute), Ramesh K. Sitaraman (UMass, Amherst & Akamai Tech)

  • Cracking Open the Black Box: What Observations Can Tell Us About Reinforcement Learning Agents

    Arnaud Dethise (KAUST), Marco Canini (KAUST), Srikanth Kandula (Microsoft)

  • Verifying Deep-RL-Driven Systems

    Yafim Kazak (The Hebrew University of Jerusalem), Clark Barrett (Stanford University), Guy Katz (The Hebrew University of Jerusalem), Michael Schapira (The Hebrew University of Jerusalem)

  • 11:30am - 13:30pm Lunch break

  • 13:30pm - 15:00pm Session 2: Analytics & Self-Driving Networks

    Session chair: Junchen Jiang (University of Chicago)

  • NetBOA: Self-Driving Network Benchmarking

    Johannes Zerwas (Technical University of Munich, Germany), Patrick Kalmbach (Technical University of Munich, Germany), Laurenz Henkel (Technical University of Munich, Germany), Gábor Rétvári (Department of Telecommunications and Media Informatics, Budapest University of Technology and Economics), Wolfgang Kellerer (Technical University of Munich, Germany), Andreas Blenk (Technical University of Munich, Germany), Stefan Schmid (Faculty of Computer Science, University of Vienna, Austria)

  • Assisting Delay and Bandwidth Sensitive Applications in a Self-Driving Network

    Sharat Chandra Madanapalli (UNSW Sydney), Hassan Habibi Gharakheili (University of New South Wales), Vijay Sivaraman (University of New South Wales)

  • ONTAS: Flexible and Scalable Online Network Traffic Anonymization System

    Hyojoon Kim (Princeton University), Arpit Gupta (Columbia University)

  • Smart Prediction of the Complaint Hotspot Problem in Mobile Network

    Lin Zhu (China Mobile Research Institute), Juan Zhao (China Mobile Research Institute), Yiting Wang (China Mobile Research Institute), Junlan Feng (China Mobile Research Institute), Chao Deng (China Mobile Research Institute), Hui Li (China Mobile Research Institute)

  • UDAAN: Embedding User-Defined Analytics Applications in Network Devices

    Anu Mercian (Hewlett Packard Labs), Puneet Sharma (Hewlett Packard Labs), Renato Aguiar (HPE Aruba), Chinlin Chen (HPE Aruba), David Pinheiro (HPE Aruba)

  • 15:00pm - 15:30pm Coffee Break

  • 15:30pm - 16:30pm Session 3: ML for Network Modeling

    Session chair: Arpit Gupta (UC Santa Barbara)

  • Towards a Profiling View for Traffic Classification by Exploring the Statistic Features and Link Patterns

    Meng Qin (ICNLAB, School of Electronics and Computer Engineering (SECE), Peking University), Kai Lei (ICNLAB, School of Electronics and Computer Engineering (SECE), Peking University), Bo Bai (Theory Lab, 2012 Labs, Huawei Technologies, Co. Ltd.), Gong Zhang (Theory Lab, 2012 Labs, Huawei Technologies, Co. Ltd.)

  • Best Paper:
    Runtime Verification of P4 Networks with Reinforcement Learning

    Apoorv Shukla (TU Berlin), Kevin Nico Hudemann (TU Berlin), Artur Hecker (Huawei Technologies), Stefan Schmid (University of Vienna)

  • Hierarchical Bayesian Modeling for Wireless Cellular Networks

    Deniz Ustebay (Huawei Noah's Ark Lab), Jie Chuai (Huawei Noah's Ark Lab)

  • 16:30pm - 17:00pm Concluding remarks

Call for Papers

In recent years, we have witnessed: (1) development of fully programmable, protocol-independent data planes and languages for programming them; and (2) the emergence of new platforms, tools, and algorithms for Artificial Intelligence (AI) and Machine Learning (ML). These technological advancements and scientific innovations create exciting new opportunities. On the one hand, the scientific innovations in the area of AI/ML have the potential to simplify network management (monitoring as well as control). On the other hand, recent advancements in the area of networking technology have the potential to improve the performance of AI and ML systems.

AI/ML for Networking. AI & ML have been successfully applied to various perceptual domains, including computer vision, natural language processing, and voice recognition. In addition, ML techniques are showing impressive results in new domains such as medicine, finance, and astronomy, to name a few. This success in non-perceptual domains suggests that ML techniques could be successfully applied to simplify network management. For at least a decade, networking researchers, equipment vendors, and Internet service providers alike have argued for “autonomous” or “self-driving” networks, where network management and control decisions are made in real time and in an automated fashion. Yet, building such “self-driving” networks that are practically deployable has largely remained unrealized. The coupling of the programmable control of software-defined networking (SDN) with scientific innovations in AI and ML promises unprecedented opportunities for querying high-volume and high-velocity, distributed streaming data at scale; this new technical capability can provide the necessary information to the many different network monitoring and control tasks that self-driving networks should perform automatically and autonomously. Building a self-driving network is one of the “grand challenges” of networking research today. Realizing this vision will require incorporating the collective expertise and input from the networking research community.

Networking for AI/ML. Distributed processing systems for Artificial Intelligence (AI) and Machine Learning (ML), such as Hadoop, Spark, Storm, GraphLab, TensorFlow etc., are widely used by industry. Networking is a well-known bottleneck for AI & ML systems. Though the recent technological advances, such as reconfigurable switches, programmable NICs, RDMA- over-converged Ethernet (RoCE) and GPU direct, etc., provide exciting opportunities to improve the performance of AI and ML solutions, but the ever-increasing complexity of networks composed of a heterogeneous set of targets makes effective monitoring, modeling, auditing, and overall control of network traffic difficult if not impossible. Hence there is a need for more powerful methods to solve the challenges faced in the design, deployment, and management of networks for distributed processing systems for AI and ML.

This workshop will provide a forum for networking researchers to present and share their latest research on building self-driving networks and coupling the technological advances in networking with scientific innovations in AI and ML. This workshop seeks contributions from experts in areas such as network programming, formal methods, control theory, distributed systems, machine learning, data science, data structures and algorithms, and optimization who share in the excitement of realizing the vision of self-driving networks as well as improving the performance of AI and ML solutions.

Topics of Interest

  • Design and implementation of systems for flexible and scalable network monitoring
  • Design and implementation of closed-loop systems (modular) that use monitoring to drive network control (e.g., congestion control, TE, QoE, QoS, etc.) with minimal human intervention
  • Unified programming languages/abstractions for expressing both network monitoring (streaming as well as offline) and control tasks
  • Algorithms to train learning models for inferring network attacks, device/service fingerprinting, congestion, failures, QoE metrics, etc. in (real time) at scale
  • New data structures, algorithms, network protocols, and switch architectures for storing and/or processing network monitoring data (single-site and/or distributed settings)
  • Query-planning algorithms to scale the execution of network-monitoring queries
  • Techniques to collect and analyze network data in a privacy-preserving manner
  • Learning models to capture the relationship between network events and control actions
  • Design data structures and algorithms for consistently and correctly updating the distributed states (e.g., forwarding table entries)
  • Examples of design choices informed by control-theoretic findings (e.g., hard limits, unavoidable tradeoffs)
  • New use cases for self-driving networks in DCs, WANs, IXPs, wireless networks, cloud networks, CDNs, home networks, etc.
  • Case studies demonstrating (dis)advantages of choosing AI/ML techniques for networking over more traditional ones
  • New topology, algorithms, network protocols, and switch architectures for AI/ML applications
  • Techniques to optimize distributed AI, ML and graph processing algorithms/systems with new networking options (e.g., PISA switches, SmartNICs, RDMA, NVLink, etc.)
  • Measurement and analysis of network traffic for AI & ML systems

Submission Instructions

Submissions must be original, unpublished work, and not under consideration at another conference or journal. Submitted papers must be at most six (6) pages long, including all figures, tables, references, and appendices in two-column 10pt ACM format. This six-page limit does not include the reproducibility section (see below). Papers must include authors names and affiliations for single-blind peer reviewing by the PC. Authors of accepted papers are expected to present their papers at the workshop. Please submit your paper via

Research Reproducibility

As the interest in coupling networking and AI/ML is growing, we recognize that reproducible work becomes ever more important, in particular, due to the somewhat inherent lack of interpretability associated with ML methods.

  • To encourage reproducibility, authors are allowed an additional but optional page in addition to the six pages. This additional space is meant to be used to explicitly discuss the reproducibility and interpretability of their solution. In general, all authors are encouraged to consult the reproducibility checklist to verify that their submission is reproducible.

  • In NetAI, we want to encourage authors to go the extra mile and release their research artifacts. We plan to reward the reproducible work with special ACM badging. All the accepted papers at the workshop are eligible for artifacts review.

Note: While NetAI encourages reproducibility, the reviewers will not place any additional preferences, while reviewing, towards papers with a reproducibility section.

Authors Take Note

The official publication date is the date the proceedings are made available in the ACM Digital Library. This date may be up to TWO WEEKS prior to the first day of the conference. The official publication date affects the deadline for any patent filings related to published work.


Attendance of the workshop is by open registration and subject to the same registration fees and rules as all the other SIGCOMM 2019 workshops. The registrants of the workshop may freely attend any workshop on the same day.

Camera-ready instructions

For the final paper to be published, please refer to Camera-ready instructions for workshops.

Important Dates

  • April 12, 2019 11:59 PST

    Paper submission deadline

  • May 17, 2019 11:59 PST

    Paper acceptance notification

  • June 20, 2019 11:59 PST

    Camera-ready deadline

  • August 23, 2019



  • General Chairs
  • Marco Canini


  • Jon Crowcroft

    University of Cambridge

  • Nick Feamster

    Princeton University

  • Jennifer Rexford

    Princeton University

  • Walter Willinger

    NIKSUN Inc.

  • Nicholas Zhang


  • Program Committee Chairs
  • Theophilus Benson

    Brown University

  • Arpit Gupta

    UC Santa Barbara

  • Junchen Jiang

    University of Chicago

  • Program Committee Members
  • Mohammad Alizadeh

    MIT, USA

  • Behnaz Arzani

    Microsoft Research, USA

  • Sujata Banerjee

    VMWare, USA

  • Marco Canini

    KAUST, Saudi Arabia

  • Aakanksha Chowdhery

    Google Brain, USA

  • Jon Crowcroft

    University of Cambridge, UK

  • Nick Feamster

    Princeton, USA

  • Chuanxiong Guo

    Bytedance, USA

  • Xin Jin

    John Hopkins Univ, USA

  • Srikanth Kandula

    Microsoft Research, USA

  • Changhoon Kim

    Barefoot, USA

  • Bryan Larish

    Verizon, USA

  • Dan Li

    Tsinghua Univ, China

  • Ihsan Ayyub Qazi

    LUMS, Pakistan

  • Matthew Roughan

    University of Adelaide, Australia

  • Rijurekha Sen

    IIT Delhi, India

  • Michael Schapira

    Hebrew University of Jerusalem, Israel

  • Chen Tian

    Nanjing University, China

  • Walter Willinger

    NIKSUN Inc., USA

  • Ying Zhang

    Facebook, USA

  • Ben Zhao

    University of Chicago, USA

  • Zhi-Li Zhang

    UMN, USA