Matches in SemOpenAlex for { <https://semopenalex.org/work/W4313443054> ?p ?o ?g. }
Showing items 1 to 73 of
73
with 100 items per page.
- W4313443054 abstract "Function approximation has enabled remarkable advances in applying reinforcement learning (RL) techniques in environments with high-dimensional inputs, such as images, in an end-to-end fashion, mapping such inputs directly to low-level control. Nevertheless, these have proved vulnerable to small adversarial input perturbations. A number of approaches for improving or certifying robustness of end-to-end RL to adversarial perturbations have emerged as a result, focusing on cumulative reward. However, what is often at stake in adversarial scenarios is the violation of fundamental properties, such as safety, rather than the overall reward that combines safety with efficiency. Moreover, properties such as safety can only be defined with respect to true state, rather than the high-dimensional raw inputs to end-to-end policies. To disentangle nominal efficiency and adversarial safety, we situate RL in deterministic partially-observable Markov decision processes (POMDPs) with the goal of maximizing cumulative reward subject to safety constraints. We then propose a partially-supervised reinforcement learning (PSRL) framework that takes advantage of an additional assumption that the true state of the POMDP is known at training time. We present the first approach for certifying safety of PSRL policies under adversarial input perturbations, and two adversarial training approaches that make direct use of PSRL. Our experiments demonstrate both the efficacy of the proposed approach for certifying safety in adversarial environments, and the value of the PSRL framework coupled with adversarial training in improving certified safety while preserving high nominal reward and high-quality predictions of true state." @default.
- W4313443054 created "2023-01-06" @default.
- W4313443054 creator A5038669899 @default.
- W4313443054 creator A5039914913 @default.
- W4313443054 creator A5056913693 @default.
- W4313443054 date "2022-12-28" @default.
- W4313443054 modified "2023-10-18" @default.
- W4313443054 title "Certifying Safety in Reinforcement Learning under Adversarial Perturbation Attacks" @default.
- W4313443054 doi "https://doi.org/10.48550/arxiv.2212.14115" @default.
- W4313443054 hasPublicationYear "2022" @default.
- W4313443054 type Work @default.
- W4313443054 citedByCount "0" @default.
- W4313443054 crossrefType "posted-content" @default.
- W4313443054 hasAuthorship W4313443054A5038669899 @default.
- W4313443054 hasAuthorship W4313443054A5039914913 @default.
- W4313443054 hasAuthorship W4313443054A5056913693 @default.
- W4313443054 hasBestOaLocation W43134430541 @default.
- W4313443054 hasConcept C104317684 @default.
- W4313443054 hasConcept C105795698 @default.
- W4313443054 hasConcept C106189395 @default.
- W4313443054 hasConcept C119857082 @default.
- W4313443054 hasConcept C121332964 @default.
- W4313443054 hasConcept C126255220 @default.
- W4313443054 hasConcept C154945302 @default.
- W4313443054 hasConcept C159886148 @default.
- W4313443054 hasConcept C163836022 @default.
- W4313443054 hasConcept C17098449 @default.
- W4313443054 hasConcept C185592680 @default.
- W4313443054 hasConcept C32848918 @default.
- W4313443054 hasConcept C33923547 @default.
- W4313443054 hasConcept C37736160 @default.
- W4313443054 hasConcept C41008148 @default.
- W4313443054 hasConcept C55493867 @default.
- W4313443054 hasConcept C62520636 @default.
- W4313443054 hasConcept C63479239 @default.
- W4313443054 hasConcept C97541855 @default.
- W4313443054 hasConcept C98763669 @default.
- W4313443054 hasConceptScore W4313443054C104317684 @default.
- W4313443054 hasConceptScore W4313443054C105795698 @default.
- W4313443054 hasConceptScore W4313443054C106189395 @default.
- W4313443054 hasConceptScore W4313443054C119857082 @default.
- W4313443054 hasConceptScore W4313443054C121332964 @default.
- W4313443054 hasConceptScore W4313443054C126255220 @default.
- W4313443054 hasConceptScore W4313443054C154945302 @default.
- W4313443054 hasConceptScore W4313443054C159886148 @default.
- W4313443054 hasConceptScore W4313443054C163836022 @default.
- W4313443054 hasConceptScore W4313443054C17098449 @default.
- W4313443054 hasConceptScore W4313443054C185592680 @default.
- W4313443054 hasConceptScore W4313443054C32848918 @default.
- W4313443054 hasConceptScore W4313443054C33923547 @default.
- W4313443054 hasConceptScore W4313443054C37736160 @default.
- W4313443054 hasConceptScore W4313443054C41008148 @default.
- W4313443054 hasConceptScore W4313443054C55493867 @default.
- W4313443054 hasConceptScore W4313443054C62520636 @default.
- W4313443054 hasConceptScore W4313443054C63479239 @default.
- W4313443054 hasConceptScore W4313443054C97541855 @default.
- W4313443054 hasConceptScore W4313443054C98763669 @default.
- W4313443054 hasLocation W43134430541 @default.
- W4313443054 hasOpenAccess W4313443054 @default.
- W4313443054 hasPrimaryLocation W43134430541 @default.
- W4313443054 hasRelatedWork W1563041104 @default.
- W4313443054 hasRelatedWork W1997350370 @default.
- W4313443054 hasRelatedWork W2096426448 @default.
- W4313443054 hasRelatedWork W2146763310 @default.
- W4313443054 hasRelatedWork W2173087131 @default.
- W4313443054 hasRelatedWork W2347690758 @default.
- W4313443054 hasRelatedWork W2953146157 @default.
- W4313443054 hasRelatedWork W3000588951 @default.
- W4313443054 hasRelatedWork W3037550388 @default.
- W4313443054 hasRelatedWork W52153049 @default.
- W4313443054 isParatext "false" @default.
- W4313443054 isRetracted "false" @default.
- W4313443054 workType "article" @default.