Interpreting Deep Learning-Based Networking Systems
Authors:
Open Access
- 30 July 2020
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 154-171
Abstract
While many deep learning (DL)-based networking systems have demonstrated superior performance, the underlying Deep Neural Networks (DNNs) remain blackboxes and stay uninterpretable for network operators. The lack of interpretability makes DL-based networking systems prohibitive to deploy in practice. In this paper, we propose Metis, a framework that provides interpretability for two general categories of networking problems spanning local and global control. Accordingly, Metis introduces two different interpretation methods based on decision tree and hypergraph, where it converts DNN policies to interpretable rule-based controllers and highlight critical components based on analysis over hypergraph. We evaluate Metis over two categories of state-of-the-art DL-based networking systems and show that Metis provides human-readable interpretations while preserving nearly no degradation in performance. We further present four concrete use cases of Metis, showcasing how Metis helps network operators to design, debug, deploy, and ad-hoc adjust DL-based networking systems.Keywords
All Related Versions
Funding Information
- National Key Research and Development Program of China (2017YFB0801701)
- National Natural Science Foundation of China (61625203, 61832013)
This publication has 44 references indexed in Scilit:
- A Control-Theoretic Approach for Dynamic Adaptive Video Streaming over HTTPACM SIGCOMM Computer Communication Review, 2015
- Deep learningNature, 2015
- Human-level control through deep reinforcement learningNature, 2015
- Commute path bandwidth traces from 3G networks2013
- A survey of intrusion detection techniques in CloudJournal of Network and Computer Applications, 2013
- Tree-Based Methods2002
- Top-down induction of first-order logical decision treesArtificial Intelligence, 1998
- Multilayer feedforward networks with a nonpolynomial activation function can approximate any functionNeural Networks, 1993
- On Information and SufficiencyThe Annals of Mathematical Statistics, 1951
- A Mathematical Theory of CommunicationBell System Technical Journal, 1948