Filters








876 Hits in 7.0 sec

When Are Linear Stochastic Bandits Attackable? [article]

Huazheng Wang, Haifeng Xu, Hongning Wang
2022 arXiv   pre-print
We study adversarial attacks on linear stochastic bandits: by manipulating the rewards, an adversary aims to control the behaviour of the bandit algorithm.  ...  This is in sharp contrast to context-free stochastic bandits, and is intrinsically due to the correlation among arms in linear stochastic bandits.  ...  Linear stochastic bandits are related to context-free stochastic bandits and linear contextual bandits.  ... 
arXiv:2110.09008v2 fatcat:umsuhse2tfdjjjph4lqeizk3uq

Observation-Free Attacks on Stochastic Bandits

Yinglun Xu, Bhuvesh Kumar, Jacob D. Abernethy
2021 Neural Information Processing Systems  
We study data corruption attacks on stochastic multi arm bandit algorithms.  ...  We further show that various popular stochastic multi arm bandit algorithms such UCB, -greedy and Thompson Sampling satisfy this sufficient condition and are thus prone to data corruption attacks.  ...  Next we introduce the notion of adversarial attacks in the stochastic bandit setting.  ... 
dblp:conf/nips/XuKA21 fatcat:3q4byzcsjfeerjun4traoul3ja

Data Poisoning Attacks on Stochastic Bandits [article]

Fang Liu, Ness Shroff
2019 arXiv   pre-print
Our adaptive attack strategy can hijack the behavior of the bandit algorithm to suffer a linear regret with only a logarithmic cost to the attacker.  ...  Our results demonstrate a significant security threat to stochastic bandits.  ...  Data Poisoning Attacks on Stochastic Bandits by (Jun et al., 2018).  ... 
arXiv:1905.06494v1 fatcat:uhmscq7d2fhmjjxse7zcm2raam

Action-Manipulation Attacks Against Stochastic Bandits: Attacks and Defense [article]

Guanlin Liu, Lifeng lai
2020 arXiv   pre-print
Due to the broad range of applications of stochastic multi-armed bandit model, understanding the effects of adversarial attacks and designing bandit algorithms robust to attacks are essential for the safe  ...  To defend against this class of attacks, we introduce a novel algorithm that is robust to action-manipulation attacks when an upper bound for the total attack cost is given.  ...  The proofs are collected in Appendix. II. MODEL In this section, we introduce our model. We consider the standard multi-armed stochastic bandit problems setting.  ... 
arXiv:2002.08000v2 fatcat:ywgcxdifwnck3mpxxjsln5zsk4

Adversarial Attacks on Linear Contextual Bandits [article]

Evrard Garcelon, Baptiste Roziere, Laurent Meunier, Jean Tarbouriech, Olivier Teytaud, Alessandro Lazaric, Matteo Pirotta
2020 arXiv   pre-print
In this paper, we study several attack scenarios and show that a malicious agent can force a linear contextual bandit algorithm to pull any desired arm T - o(T) times over a horizon of T steps, while applying  ...  Contextual bandit algorithms are applied in a wide range of domains, from advertising to recommender systems, from clinical trials to education.  ...  We observe empirically that the total cost of attack is sublinear when using r 2 . [13] does not assume that rewards are bounded but focus on attacking algorithms in the stochastic multi-armed setting.  ... 
arXiv:2002.03839v3 fatcat:oweqjzh4erh7pfosmss2sovvcm

Adversarial Attacks on Stochastic Bandits [article]

Kwang-Sung Jun, Lihong Li, Yuzhe Ma, Xiaojin Zhu
2018 arXiv   pre-print
We study adversarial attacks that manipulate the reward signals to control the actions chosen by a stochastic multi-armed bandit algorithm.  ...  As bandits are seeing increasingly wide use in practice, our study exposes a significant security threat.  ...  [21] considers attacking stochastic contextual bandit algorithms.  ... 
arXiv:1810.12188v1 fatcat:kc5dt6pvb5g65aidbxw7botyze

Stochastic Linear Bandits Robust to Adversarial Attacks [article]

Ilija Bogunovic, Arpan Losalka, Andreas Krause, Jonathan Scarlett
2020 arXiv   pre-print
We consider a stochastic linear bandit problem in which the rewards are not only subject to random noise, but also adversarial attacks subject to a suitable budget C (i.e., an upper bound on the sum of  ...  Both variants are shown to attain near-optimal regret in the non-corrupted case C = 0, while incurring additional additive terms respectively having a linear and quadratic dependency on C in general.  ...  [20] also study stochastic linear bandits with adversarial corruptions.  ... 
arXiv:2007.03285v2 fatcat:66mro6ndwfaqdojz4rmplsq324

Near Optimal Adversarial Attack on UCB Bandits [article]

Shiliang Zuo
2022 arXiv   pre-print
We consider a stochastic multi-arm bandit problem where rewards are subject to adversarial corruption.  ...  We also prove the first lower bound on the cumulative attack cost. Our lower bound matches our upper bound up to loglog T factors, showing our attack to be near optimal.  ...  Premilinaries We consider a stochastic multi-arm bandit problem where rewards are subject to adversarial corruptions. Let T be the time horizon and K the number of arms.  ... 
arXiv:2008.09312v2 fatcat:45jrkuidxvfgbolvkdfw6n6nvi

Efficient Action Poisoning Attacks on Linear Contextual Bandits [article]

Guanlin Liu, Lifeng Lai
2021 arXiv   pre-print
We design action poisoning attack schemes against linear contextual bandit algorithms in both white-box and black-box settings.  ...  In order to develop trustworthy contextual bandit systems, understanding the impacts of various adversarial attacks on contextual bandit algorithms is essential.  ...  In the linear contextual bandit setting, [31] proposes a stochastic linear bandit algorithm, called Robust Phased Elimination (RPE), that is robust to reward poisoning attacks.  ... 
arXiv:2112.05367v1 fatcat:3frbcvuz3za7pisisqnsyvpz4u

Adversarial Attacks on Gaussian Process Bandits [article]

Eric Han, Jonathan Scarlett
2022 arXiv   pre-print
Our goal is to understand adversarial attacks on GP bandits from theoretical and practical perspectives.  ...  Recent works on GP bandits have proposed to move beyond random noise and devise algorithms robust to adversarial attacks.  ...  arXiv:2110.08449v3 [stat.ML] 16 Jun 2022 Our study is related to that of attacks on stochastic linear bandits (Garcelon et al., 2020) , but we move to the GP setting which is inherently non-linear and  ... 
arXiv:2110.08449v3 fatcat:45sjlxok3vaxblxwicarkytcea

Saving Stochastic Bandits from Poisoning Attacks via Limited Data Verification [article]

Anshuka Rangi, Long Tran-Thanh, Haifeng Xu, Massimo Franceschetti
2022 arXiv   pre-print
We study bandit algorithms under data poisoning attacks in a bounded reward setting.  ...  On the other hand, when the number of verifications is bounded above by a budget B, we propose a novel algorithm, Secure-BARBAR, which provably achieves O(min{C,T/√(B)}) regret with high probability against  ...  Stochastic linear bandits robust to adversarial attacks. arXiv preprint arXiv:2007.03285, 2020. Thodoris Lykouris, Vahab Mirrokni, and Renato Paes Leme.  ... 
arXiv:2102.07711v2 fatcat:5sbw5gqvxbcm5e7wsdg3by7phm

Saving Stochastic Bandits from Poisoning Attacks via Limited Data Verification

Anshuka Rangi, Long Tran-Thanh, Haifeng Xu, Massimo Franceschetti
2022 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
This paper studies bandit algorithms under data poisoning attacks in a bounded reward setting.  ...  against weak attackers (i.e., attackers who have to place the contamination before seeing the actual pulls of the bandit algorithm), where C is the total amount of contamination by the attacker, which  ...  Therefore, to guarantee sub-linear regret when the attacker has an unbounded amount of contamination it is necessary for the bandit algorithm to exploit additional (and possibly costly) resources.  ... 
doi:10.1609/aaai.v36i7.20777 fatcat:c3cdf7s7i5fnhb5mq3tixtyyge

Robust Stochastic Bandit Algorithms under Probabilistic Unbounded Adversarial Attack

Ziwei Guan, Kaiyi Ji, Donald J. Bucci Jr., Timothy Y. Hu, Joseph Palombo, Michael Liston, Yingbin Liang
2020 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
Both of these algorithms are provably robust to the aforementioned attack model.  ...  The multi-armed bandit formalism has been extensively studied under various attack models, in which an adversary can modify the reward revealed to the player.  ...  and O(log T ) for stochastic bandit without attack.  ... 
doi:10.1609/aaai.v34i04.5821 fatcat:se34f6ghfjgvrk3okzznuz2nnm

Online Data Poisoning Attack [article]

Xuezhou Zhang, Xiaojin Zhu, Laurent Lessard
2019 arXiv   pre-print
We formulate online data poisoning attack as a stochastic optimal control problem, and solve it with model predictive control and deep reinforcement learning.  ...  We study data poisoning attacks in the online setting where training items arrive sequentially, and the attacker may perturb the current item to manipulate online learning.  ...  A variety of attacks against other learners have been developed, including neural networks [20, 12] , autoregressive models [7, 1] , linear and stochastic bandits [17, 11] , collaborative filtering  ... 
arXiv:1903.01666v2 fatcat:fimail4tyvgwnnra7dyc5i23sa

Robust Stochastic Bandit Algorithms under Probabilistic Unbounded Adversarial Attack [article]

Ziwei Guan, Kaiyi Ji, Donald J Bucci Jr, Timothy Y Hu, Joseph Palombo, Michael Liston, Yingbin Liang
2020 arXiv   pre-print
Both of these algorithms are provably robust to the aforementioned attack model.  ...  The multi-armed bandit formalism has been extensively studied under various attack models, in which an adversary can modify the reward revealed to the player.  ...  and O(log T ) for stochastic bandit without attack.  ... 
arXiv:2002.07214v1 fatcat:jp4qtiig45d5rii4ueppstfe2q
« Previous Showing results 1 — 15 out of 876 results