Matches in SemOpenAlex for { <https://semopenalex.org/work/W3005360685> ?p ?o ?g. }
Showing items 1 to 92 of
92
with 100 items per page.
- W3005360685 abstract "This chapter studies emerging cyber-attacks on reinforcement learning (RL) and introduces a quantitative approach to analyze the vulnerabilities of RL. Focusing on adversarial manipulation on the cost signals, we analyze the performance degradation of TD($lambda$) and $Q$-learning algorithms under the manipulation. For TD($lambda$), the approximation learned from the manipulated costs has an approximation error bound proportional to the magnitude of the attack. The effect of the adversarial attacks on the bound does not depend on the choice of $lambda$. In $Q$-learning, we show that $Q$-learning algorithms converge under stealthy attacks and bounded falsifications on cost signals. We characterize the relation between the falsified cost and the $Q$-factors as well as the policy learned by the learning agent which provides fundamental limits for feasible offensive and defensive moves. We propose a robust region in terms of the cost within which the adversary can never achieve the targeted policy. We provide conditions on the falsified cost which can mislead the agent to learn an adversary's favored policy. A case study of TD($lambda$) learning is provided to corroborate the results." @default.
- W3005360685 created "2020-02-14" @default.
- W3005360685 creator A5060454132 @default.
- W3005360685 creator A5081500464 @default.
- W3005360685 date "2020-02-07" @default.
- W3005360685 modified "2023-09-24" @default.
- W3005360685 title "Manipulating Reinforcement Learning: Poisoning Attacks on Cost Signals" @default.
- W3005360685 cites W1515851193 @default.
- W3005360685 cites W1977655452 @default.
- W3005360685 cites W2071983464 @default.
- W3005360685 cites W2073320424 @default.
- W3005360685 cites W2074500080 @default.
- W3005360685 cites W2082435755 @default.
- W3005360685 cites W2098172344 @default.
- W3005360685 cites W2107221565 @default.
- W3005360685 cites W2137766593 @default.
- W3005360685 cites W2145339207 @default.
- W3005360685 cites W2150878689 @default.
- W3005360685 cites W2163468230 @default.
- W3005360685 cites W2288800863 @default.
- W3005360685 cites W2895478303 @default.
- W3005360685 cites W2896016186 @default.
- W3005360685 cites W2952603690 @default.
- W3005360685 cites W2954035548 @default.
- W3005360685 cites W2963943581 @default.
- W3005360685 cites W2970912396 @default.
- W3005360685 cites W2979429887 @default.
- W3005360685 cites W2981396729 @default.
- W3005360685 cites W3006912862 @default.
- W3005360685 cites W3011120880 @default.
- W3005360685 doi "https://doi.org/10.48550/arxiv.2002.03827" @default.
- W3005360685 hasPublicationYear "2020" @default.
- W3005360685 type Work @default.
- W3005360685 sameAs 3005360685 @default.
- W3005360685 citedByCount "0" @default.
- W3005360685 crossrefType "posted-content" @default.
- W3005360685 hasAuthorship W3005360685A5060454132 @default.
- W3005360685 hasAuthorship W3005360685A5081500464 @default.
- W3005360685 hasBestOaLocation W30053606851 @default.
- W3005360685 hasConcept C119857082 @default.
- W3005360685 hasConcept C120665830 @default.
- W3005360685 hasConcept C121332964 @default.
- W3005360685 hasConcept C127413603 @default.
- W3005360685 hasConcept C134306372 @default.
- W3005360685 hasConcept C154945302 @default.
- W3005360685 hasConcept C176217482 @default.
- W3005360685 hasConcept C176856949 @default.
- W3005360685 hasConcept C21547014 @default.
- W3005360685 hasConcept C2778113609 @default.
- W3005360685 hasConcept C33923547 @default.
- W3005360685 hasConcept C34388435 @default.
- W3005360685 hasConcept C37736160 @default.
- W3005360685 hasConcept C38652104 @default.
- W3005360685 hasConcept C41008148 @default.
- W3005360685 hasConcept C41065033 @default.
- W3005360685 hasConcept C42475967 @default.
- W3005360685 hasConcept C97541855 @default.
- W3005360685 hasConceptScore W3005360685C119857082 @default.
- W3005360685 hasConceptScore W3005360685C120665830 @default.
- W3005360685 hasConceptScore W3005360685C121332964 @default.
- W3005360685 hasConceptScore W3005360685C127413603 @default.
- W3005360685 hasConceptScore W3005360685C134306372 @default.
- W3005360685 hasConceptScore W3005360685C154945302 @default.
- W3005360685 hasConceptScore W3005360685C176217482 @default.
- W3005360685 hasConceptScore W3005360685C176856949 @default.
- W3005360685 hasConceptScore W3005360685C21547014 @default.
- W3005360685 hasConceptScore W3005360685C2778113609 @default.
- W3005360685 hasConceptScore W3005360685C33923547 @default.
- W3005360685 hasConceptScore W3005360685C34388435 @default.
- W3005360685 hasConceptScore W3005360685C37736160 @default.
- W3005360685 hasConceptScore W3005360685C38652104 @default.
- W3005360685 hasConceptScore W3005360685C41008148 @default.
- W3005360685 hasConceptScore W3005360685C41065033 @default.
- W3005360685 hasConceptScore W3005360685C42475967 @default.
- W3005360685 hasConceptScore W3005360685C97541855 @default.
- W3005360685 hasLocation W30053606851 @default.
- W3005360685 hasOpenAccess W3005360685 @default.
- W3005360685 hasPrimaryLocation W30053606851 @default.
- W3005360685 hasRelatedWork W2604394466 @default.
- W3005360685 hasRelatedWork W2941205169 @default.
- W3005360685 hasRelatedWork W2952603690 @default.
- W3005360685 hasRelatedWork W2955689724 @default.
- W3005360685 hasRelatedWork W3005360685 @default.
- W3005360685 hasRelatedWork W3038682671 @default.
- W3005360685 hasRelatedWork W3108220207 @default.
- W3005360685 hasRelatedWork W3126884055 @default.
- W3005360685 hasRelatedWork W3176644864 @default.
- W3005360685 hasRelatedWork W4311943648 @default.
- W3005360685 isParatext "false" @default.
- W3005360685 isRetracted "false" @default.
- W3005360685 magId "3005360685" @default.
- W3005360685 workType "article" @default.