FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3037974410> ?p ?o ?g. }

• W3037974410 abstract "Machine learning systems are deployed in critical settings, but they might fail in unexpected ways, impacting the accuracy of their predictions. Poisoning attacks against machine learning induce adversarial modification of data used by a machine learning algorithm to selectively change its output when it is deployed. In this work, we introduce a novel data poisoning attack called a emph{subpopulation attack}, which is particularly relevant when datasets are large and diverse. We design a modular framework for subpopulation attacks, instantiate it with different building blocks, and show that the attacks are effective for a variety of datasets and machine learning models. We further optimize the attacks in continuous domains using influence functions and gradient optimization methods. Compared to existing backdoor poisoning attacks, subpopulation attacks have the advantage of inducing misclassification in naturally distributed data points at inference time, making the attacks extremely stealthy. We also show that our attack strategy can be used to improve upon existing targeted attacks. We prove that, under some assumptions, subpopulation attacks are impossible to defend against, and empirically demonstrate the limitations of existing defenses against our attacks, highlighting the difficulty of protecting machine learning against this threat." @default.
• W3037974410 created "2020-07-02" @default.
• W3037974410 creator A5014250337 @default.
• W3037974410 creator A5035574749 @default.
• W3037974410 creator A5054655342 @default.
• W3037974410 creator A5054966546 @default.
• W3037974410 date "2020-06-24" @default.
• W3037974410 modified "2023-09-27" @default.
• W3037974410 title "Subpopulation Data Poisoning Attacks" @default.
• W3037974410 cites W1552056088 @default.
• W3037974410 cites W1686810756 @default.
• W3037974410 cites W2007562169 @default.
• W3037974410 cites W2033368661 @default.
• W3037974410 cites W2038296020 @default.
• W3037974410 cites W2064675550 @default.
• W3037974410 cites W2100960835 @default.
• W3037974410 cites W2101234009 @default.
• W3037974410 cites W2105037940 @default.
• W3037974410 cites W2108598243 @default.
• W3037974410 cites W2112507308 @default.
• W3037974410 cites W2113459411 @default.
• W3037974410 cites W2165698076 @default.
• W3037974410 cites W2167421362 @default.
• W3037974410 cites W2293844262 @default.
• W3037974410 cites W2530395818 @default.
• W3037974410 cites W2535690855 @default.
• W3037974410 cites W2592232824 @default.
• W3037974410 cites W2597603852 @default.
• W3037974410 cites W2603766943 @default.
• W3037974410 cites W2748789698 @default.
• W3037974410 cites W2772825438 @default.
• W3037974410 cites W2774423163 @default.
• W3037974410 cites W2788481061 @default.
• W3037974410 cites W2795435272 @default.
• W3037974410 cites W2799420851 @default.
• W3037974410 cites W2804935296 @default.
• W3037974410 cites W2807363941 @default.
• W3037974410 cites W2898998737 @default.
• W3037974410 cites W2900018096 @default.
• W3037974410 cites W2900120080 @default.
• W3037974410 cites W2916360674 @default.
• W3037974410 cites W2934843808 @default.
• W3037974410 cites W2942091739 @default.
• W3037974410 cites W2962763344 @default.
• W3037974410 cites W2963058500 @default.
• W3037974410 cites W2963207607 @default.
• W3037974410 cites W2963262394 @default.
• W3037974410 cites W2963341956 @default.
• W3037974410 cites W2963343288 @default.
• W3037974410 cites W2963403868 @default.
• W3037974410 cites W2963777610 @default.
• W3037974410 cites W2963857521 @default.
• W3037974410 cites W2964041528 @default.
• W3037974410 cites W2964135521 @default.
• W3037974410 cites W2964153729 @default.
• W3037974410 cites W2964301649 @default.
• W3037974410 cites W2967540978 @default.
• W3037974410 cites W2970597249 @default.
• W3037974410 cites W2970631161 @default.
• W3037974410 cites W2989835735 @default.
• W3037974410 cites W2995525544 @default.
• W3037974410 cites W3007437825 @default.
• W3037974410 cites W3017348803 @default.
• W3037974410 cites W3035261884 @default.
• W3037974410 cites W3035729345 @default.
• W3037974410 cites W3037024761 @default.
• W3037974410 cites W3037144731 @default.
• W3037974410 cites W3046527848 @default.
• W3037974410 cites W3101427066 @default.
• W3037974410 cites W3106646114 @default.
• W3037974410 cites W3107337211 @default.
• W3037974410 cites W3116515605 @default.
• W3037974410 cites W3118608800 @default.
• W3037974410 cites W3120223105 @default.
• W3037974410 cites W3120740533 @default.
• W3037974410 cites W3128839796 @default.
• W3037974410 cites W3131061281 @default.
• W3037974410 cites W3153022867 @default.
• W3037974410 cites W9657784 @default.
• W3037974410 hasPublicationYear "2020" @default.
• W3037974410 type Work @default.
• W3037974410 sameAs 3037974410 @default.
• W3037974410 citedByCount "10" @default.
• W3037974410 countsByYear W30379744102020 @default.
• W3037974410 countsByYear W30379744102021 @default.
• W3037974410 countsByYear W30379744102022 @default.
• W3037974410 crossrefType "posted-content" @default.
• W3037974410 hasAuthorship W3037974410A5014250337 @default.
• W3037974410 hasAuthorship W3037974410A5035574749 @default.
• W3037974410 hasAuthorship W3037974410A5054655342 @default.
• W3037974410 hasAuthorship W3037974410A5054966546 @default.
• W3037974410 hasConcept C101468663 @default.
• W3037974410 hasConcept C111919701 @default.
• W3037974410 hasConcept C119857082 @default.
• W3037974410 hasConcept C154945302 @default.
• W3037974410 hasConcept C2776214188 @default.
• W3037974410 hasConcept C2778403875 @default.
• W3037974410 hasConcept C2781045450 @default.
• W3037974410 hasConcept C37736160 @default.
• W3037974410 hasConcept C38652104 @default.

• next