Robustifying Network Protocols with Adversarial Examples

Gilad, Tomer; Jay, Nathan; Shnaiderman, Michael; Godfrey, Brighten; Schapira, Michael

doi:10.1145/3365609.3365862

Cited by 8 publications

(11 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(Our environment generator and a full list of parameters are documented in §A.2.) In recent papers, both trace-driven (e.g., [19,31]) and synthetic environments (e.g., [24,30]) are used to train RL-based network algorithms. We will explain in §4.2 how our technique applies to both types of environments.…”

Section: Motivationmentioning

confidence: 99%

See 1 more Smart Citation

Automatic Curriculum Generation for Learning Adaptation in Networking

Xia¹,

Zhou²,

Yan³

et al. 2022

Preprint

View full text Add to dashboard Cite

As deep reinforcement learning (RL) showcases its strengths in networking and systems, its pitfalls also come to the public's attention-when trained to handle a wide range of network workloads and previously unseen deployment environments, RL policies often manifest suboptimal performance and poor generalizability.To tackle these problems, we present Genet, a new training framework for learning better RL-based network adaptation algorithms. Genet is built on the concept of curriculum learning, which has proved effective against similar issues in other domains where RL is extensively employed. At a high level, curriculum learning gradually presents more difficult environments to the training, rather than choosing them randomly, so that the current RL model can make meaningful progress in training. However, applying curriculum learning in networking is challenging because it remains unknown how to measure the "difficulty" of a network environment.Instead of relying on handcrafted heuristics to determine the environment's difficulty level, our insight is to utilize traditional rule-based (non-RL) baselines: If the current RL model performs significantly worse in a network environment than the baselines, then the model's potential to improve when further trained in this environment is substantial. Therefore, Genet automatically searches for the environments where the current model falls significantly behind a traditional baseline scheme and iteratively promotes these environments as the training progresses. Through evaluating Genet on three use cases-adaptive video streaming, congestion control, and load balancing, we show that Genet produces RL policies which outperform both regularly trained RL policies and traditional baselines in each context, not only under synthetic workloads but also in real environments.

show abstract

Section: Motivationmentioning

confidence: 99%

“…• First, they use innate properties of each environment (e.g., shorter network or workload traces [33] and smoother network conditions [19] are supposedly easier), but these innate properties fail to indicate whether the current RL model can be improved in an environment.…”

Section: Introductionmentioning

confidence: 99%

Automatic Curriculum Generation for Learning Adaptation in Networking

Xia¹,

Zhou²,

Yan³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Our work builds on top of this work, and we perform an in-depth analysis of two different controllers. [10] focuses on robustness properties for Pensieve and proposes a new training procedure to enhance the robustness of the network controller. Finally, our work on finding decision boundaries is based on the reachability analysis work by [31] as we perform an enumeration of feasible RELU assignments.…”

Section: Related Workmentioning

confidence: 99%

“…However, a challenge in applying ML in networking applications is that a NN is like a black box: it takes an input and produces an output, but does not offer any insight into why that output was chosen over other possible choices. This causes a lack of assurances and trust about system behavior, which could respond in unexpected or incorrect ways [9], [34], and overall hinders the adoption of learning-based solutions.…”

Section: Introductionmentioning

confidence: 99%

Analyzing Learning-Based Networked Systems with Formal Verification

Dethise

Canini

Narodytska

2021

IEEE INFOCOM 2021 - IEEE Conference on Computer Communications

View full text Add to dashboard Cite

show abstract

“…Against these challenges, the need for fast, inexpensive, and realistic evaluation (using ML or statistical models [10,12]) of network protocols is only growing; more so with the emerging interest in using ML, including reinforcement learning, to learn or tune protocols [18,21,25,32,40].…”

Section: Related Workmentioning

confidence: 99%