Skip to content
Explaining Deep Learning-Based Network Systems
JavaScript Python C++ HTML CSS Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
case_1 Update all readmes. Oct 3, 2019
case_2 Specify the version of tf.js to 0.14.1 Dec 14, 2019
case_3 Clear up directories. Nov 1, 2019
explain_auto Update all readmes. Oct 3, 2019
explain_pensieve Fix typos. Dec 7, 2019
LICENSE Initial commit Sep 14, 2019
TranSys-TR.pdf Upload Technical Report. Dec 2, 2019

TranSys: Explaining Deep Learning-Based Networked Systems

TranSys is an integrated explainer to provide post-hoc explanations for different types of Deep Learning (DL)-based networked systems. We refer the readers to TranSys-TR.pdf for the technical report of this project.

In the current stage, we provide the explanation methods and implementations for three DL-based networked systems:

  • Pensieve (explain_pensieve) is an adaptive video streaming algorithm based on deep reinforcement learning.
  • AuTO (explain_auto) is an on-switch traffic scheduler in datacenters (under refactoring).
  • RouteNet (explain_routenet) is an SDN traffic optimizer to find routes for all src-dst pairs.

We further provide several use cases of TranSys:

  • We troubleshoot the DNN in Pensieve and improve the average quality of experience (QoE) by up to 3% over DNN policies with only decision trees (case_1).
  • With decision trees generated by TranSys, we lightweightify Pensieve and AuTO and achieve shorter decision-making latency by 27x on average and lower resource consumption by up to 156x (case_2).
  • We also provide an efficient way to compare the latency of several paths in traffic optimization based on the explanations provided by TranSys (case_3).

The running scripts for explanation methods and use cases could be found in respective directories. Currently we are still working on documentating and refactoring the repository. Other codes will be released very soon.

For any questions, please post an issue or send an email to

Release Progress

Directory Date
explain_pensieve Sep-22-2019
explain_auto In progress
explain_routenet Sep-23-2019
case_1 In progress
case_2 Nov-20-2019
case_3 Oct-31-2019

We anticipate all codes to be released soon. Please stay tuned!


  title =   {Explaining Deep Learning-Based Networked Systems},
  author =  {Meng, Zili and Wang, Minhu and Xu, Mingwei and Mao, Hongzi and Bai, Jiasong and Hu, Hongxin},
  journal = {arXiv preprint arXiv:1910.03835},
  pdf =     {},
  url =     {},
  year =    {2019}
You can’t perform that action at this time.