RECORD
Robust Deep Reinforcement Learning through Bootstrapped Opportunistic Curriculum
Machine Learning · Submitted 21 Jun 2022
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim In our experiments we show that the proposed BCL framework enables dramatic improvements in robustness of learned policies to adversarial perturbations.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Despite considerable advances in deep reinforcement learning, it has been shown to be highly vulnerable to adversarial perturbations to state observations. Recent efforts that have attempted to improve adversarial robustness of reinforcement learning can nevertheless tolerate only very small perturbations, and remain fragile as perturbation size increases. We propose Bootstrapped Opportunistic Adversarial Curriculum Learning (BCL), a novel flexible adversarial curriculum learning framework for robust reinforcement learning. Our framework combines two ideas: conservatively bootstrapping each curriculum phase with highest quality solutions obtained from multiple runs of the previous phase, and opportunistically skipping forward in the curriculum. In our experiments we show that the proposed BCL framework enables dramatic improvements in robustness of learned policies to adversarial perturbations. The greatest improvement is for Pong, where our framework yields robustness to perturbations of up to 25/255; in contrast, the best existing approach can only tolerate adversarial noise up to 5/255. Our code is available at: https://github.com/jlwu002/BCL.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Despite considerable advances in deep reinforcement learning, it has been shown to be highly vulnerable to adversarial perturbations to state observations. | background 0.99 | 0.07 | 0.02 | asserted 1.00 |
| s1 | Recent efforts that have attempted to improve adversarial robustness of reinforcement learning can nevertheless tolerate only very small perturbations, and remain fragile as perturbation size increases. | gap 0.60 | 0.05 | 0.02 | asserted 1.00 |
| s2 | We propose Bootstrapped Opportunistic Adversarial Curriculum Learning (BCL), a novel flexible adversarial curriculum learning framework for robust reinforcement learning. | contribution 0.98 | 0.82 | 0.10 | asserted 0.99 |
| s3 | Our framework combines two ideas: conservatively bootstrapping each curriculum phase with highest quality solutions obtained from multiple runs of the previous phase, and opportunistically skipping forward in the curriculum. | method 0.96 | 0.71 | 0.04 | asserted 0.97 |
| s4 | In our experiments we show that the proposed BCL framework enables dramatic improvements in robustness of learned policies to adversarial perturbations. | result 0.99 | 0.96 | 0.68 | asserted 1.00 |
| s5 | The greatest improvement is for Pong, where our framework yields robustness to perturbations of up to 25/255; in contrast, the best existing approach can only tolerate adversarial noise up to 5/255. | result 1.00 | 0.96 | 0.35 | asserted 1.00 |
| s6 | Our code is available at: https://github.com/jlwu002/BCL. | contribution 0.61 | 0.42 | 0.01 | asserted 0.86 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Despite considerable advances in deep reinforcement learning, it has been shown to be highly vulnerable to adversarial perturbations to state observations. Recent efforts that have attempted to improve adversarial robustness of reinforcement learning can nevertheless tolerate only very small perturbations, and remain fragile as perturbation size increases. We propose Bootstrapped Opportunistic Adversarial Curriculum Learning (BCL), a novel flexible adversarial curriculum learning framework for robust reinforcement learning. Our framework combines two ideas: conservatively bootstrapping each curriculum phase with highest quality solutions obtained from multiple runs of the previous phase, and opportunistically skipping forward in the curriculum. In our experiments we show that the proposed BCL framework enables dramatic improvements in robustness of learned policies to adversarial perturbations. The greatest improvement is for Pong, where our framework yields robustness to perturbations of up to 25/255; in contrast, the best existing approach can only tolerate adversarial noise up to 5/255. Our code is available at: https://github.com/jlwu002/BCL.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 4 sentences without a confident label
Contests 1 paper
-
In particular, both [ref] and [ref] propose to construct a simple curriculum of increasing input difficulty; this is what we call naive curriculum learning below, and we show that it is not particularly effective in achieving robustness in DRL.
Instance adaptive adversarial training: Improved accuracy tradeoffs in neural nets · Introduction · contest 0.79
Cites as background 5 papers
-
Studies of adversarial state perturbations to policies learned using deep reinforcement learning are somewhat more recent [ref] , as are approaches for increasing robustness [ref] .
Delving into adversarial attacks on deep policies · Related Work · 0.94
-
Studies of adversarial state perturbations to policies learned using deep reinforcement learning are somewhat more recent [ref] , as are approaches for increasing robustness [ref] .
Noisy Networks for Exploration · Related Work · 0.94
-
Studies of adversarial state perturbations to policies learned using deep reinforcement learning are somewhat more recent [ref] , as are approaches for increasing robustness [ref] .
Whatever Does Not Kill Deep Reinforcement Learning, Makes It Stronger · Related Work · 0.94
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.