Seminar in Communication Networks:
Learning, Reasoning and Control
Fall 2019

In this seminar participating students review, present, and discuss (mostly recent) research papers in the area of computer networks. During the fall semester of 2019, the seminar will focus on topics blending networks with machine learning and control theory.

We made all the materials used during the lecture public. Check out our github repository

Enrolment is open. The number of places is limited, register now!

News

May 14 Website with tentative schedule for 2019 goes live.

Tentative timeline

Contact

Professor: Laurent Vanbever ()

Research group: Networked Systems

Assistants:

Location & time

Lecture: Wednesday 13 pm–15 pm in ETZ K 91

Objectives

The two main goals of this seminar are:

  • Learning how to read and review scientific papers.
  • Learning how to present and discuss technical topics with an audience of peers.

Students are required to attend the entire seminar, choose a paper to present from a given list, prepare and give a presentation on that topic, and lead the follow-up discussion. To ensure the talks' quality, each student will be mentored by a teaching assistant. In addition to presenting one paper, every student is also required to submit one (short) review for one of the two papers presented every week in-class (12 reviews in total).

The students will be evaluated based on their submitted reviews, their presentation, their leadership in animating the discussion for their own paper, and their participation in the discussions of other papers.

Content

The seminar will start with two introductory lectures in week 1 and week 2. Starting from week 3, participating students will start reviewing, presenting, and discussing research papers. Each week will see two presentations, for a total of 24 papers.

The course content will vary from semester to semester. During the fall semester of 2019, the seminar will focus on topics blending networks with machine learning and control theory.

Prerequisites / Notice

  • Communication Networks (227-0120-00L), or equivalents.
  • Students are expected to have prior knowledge in machine learning and control theory, for instance by having attended appropriate courses.

Performance assessment

  • ECTS credits: 2 credits.
  • Type: Graded semester performance.
Part 1
Introduction
Week 1
Sept 18 Session 1 Introduction to the Course (Part 1)
Materials
Week 2
Sept 25 Session 2 Introduction to the Course (Part 2)
Materials
  • Slides (soon)
  • Handout (soon)
Exercise

Before this session, you should read the paper: Machine Learning for Networking: Workflow, Advances and Opportunities.

Week 3
Oct 2 Session 3 Overview
Materials
  • An Experimental Study of the Learnability of Congestion Control: Paper
Interesting (optional) readings
  • Congestion Control Throwdown: Paper
  • Knowledge-Defined Networking: Paper
  • A Comprehensive Survey on Machine Learning for Networking: Evolution, Applications and Research Opportunities: Paper
Review 1

Please, submit your reviews before each session.
Detailed instructions will be provided here (soon) (use your nethz credentials to sign in).

If you don't have a laptop, please let us know during the first session.

Part 2
Network Perspective
Week 4
Oct 9 Session 4 Network Measurements: Traffic Analysis and Classification
Materials
  • Neural Packet Classification: Paper
  • Sibyl: A Practical Internet Route Oracle: Paper
Interesting (optional) readings
  • Realtime Classification for Encrypted Traffic: Paper
  • Deep Learning for Encrypted Traffic Classification: An Overview: Paper
Week 5
Oct 16 Session 5 Network Measurements: Anomaly Detection
Materials
  • Outside the Closed World: On Using ML for Network Intrusion Detection: Paper
  • Detecting Credential Spearphishing Attacks in Enterprise Settings: Paper
Interesting (optional) readings
  • Demystifying Deep Learning in Networking: Paper
  • On the Effectiveness of Machine and Deep Learning for Cyber Security: Paper
Week 6
Oct 23 Session 6 Network Configuration
Materials
  • Routing or Computing? The Paradigm Shift Towards Intelligent Computer Network Packet Transmission Based on Deep Learning: Paper
  • CherryPick: Adaptively Unearthing the Best Cloud Configurations
    for BigData Analytics: Paper
Interesting (optional) readings
  • Learning to Route: Paper
  • DeepConfig: Automating Data Center Network Topologies Management with Machine Learning: Paper
  • Configtron: Tackling Network Diversity with Heterogeneous Configurations: Paper
Week 7
Oct 30 Session 7 Network Adaptation
Materials
  • AuTO: Scaling Deep Reinforcement Learning for Datacenter-Scale Automatic Traffic Optimization: Paper
  • Learning Scheduling Algorithms for Data Processing Clusters: Paper
Interesting (optional) readings
  • Why (and How) Networks Should Run Themselves: Paper
  • Resource Management with Deep Reinforcement Learning: Paper
Part 3
End-host Perspective
Week 8
Nov 6 Session 8 Congestion Control: Models and Analysis
Materials
  • On Designing Improved Controllers for AQM Routers Supporting TCP Flows: Paper
  • Rate Control for Communication Networks: Shadow Prices, Proportional Fairness and Stability: Paper
Week 9
Nov 13 Session 9 Congestion Control: Learning Algorithms
Materials
  • TCP ex Machina: Computer-Generated Congestion Control: Paper
  • PCC Vivace: Online-Learning Congestion Control: Paper
Week 10
Nov 20 Session 10 Application-level Adaptation (Part 1)
Materials
  • CS2P: Improving Video Bitrate Selection and Adaptation with Data-Driven Throughput Prediction: Paper
  • Neural Adaptive Video Streaming with Pensieve: Paper
Week 11
Nov 27 Session 11 Application-level Adaptation (Part 2)
Materials
  • AWStream: Adaptive Wide-Area Streaming Analytics: Paper
  • Pytheas: Enabling Data-Driven Quality of Experience Optimization Using Group-Based Exploration-Exploitation: Paper
Interesting (optional) readings
  • CFA: A Practical Prediction System for Video QoE Optimization: Paper
  • Inferring Netflix User Experience from Broadband Network Measurement: Paper
Part 4
New Directions
Week 12
Dec 4 Session 12 In-Network Machine Learning
Materials
  • Scaling Distributed Machine Learning with In-Network Aggregation: Paper
Interesting (optional) readings
  • Can the Network be the AI Accelerator?: Paper
  • In-network Neural Networks: Paper
Week 13—14
Dec 11 — Dec 18 Session 13 — 14 Machine-Learning Tools and Techniques
Interesting (optional) readings
  • Bringing a GAN to a Knife-fight: Adapting Malware Communication to Avoid Detection: Paper
  • Flow-based Network Traffic Generation using Generative Adversarial Networks: Paper