Matches in SemOpenAlex for { <https://semopenalex.org/work/W4229011569> ?p ?o ?g. }
- W4229011569 abstract "Training data may include personal information such as human faces, which requires anonymization to provide user privacy. However, after anonymization, the performance of the original machine learning (ML) model degrades due to the reduced or missing information. In this work, we introduce a novel privacy-preserving tensor decomposition (PTD) method to anonymize human faces. Further, we evaluate real vs. fake human face detection task as a practical use case scenario. Our approach achieves high performance as well as training data efficiency, where the essence of our approach is based on tensor decomposition to ensure face data privacy. In particular, we demonstrate that the core tensor of Tucker decomposition generated from the original face input can effectively represent the underlying characteristics of the original face data; that is, learning only from the core tensors is sufficient for differentiating real human face images from deepfakes. Also, we show that the original human face inputs are anonymized and cannot be recovered from the core tensors under different attacker models from the randomized HOOI algorithm. Through extensive experiments and analysis, we demonstrate that our method can result in high detection performance comparable to those of popular anonymization methods. Therefore, we show that our work strikes the balance between privacy and performance through the novel use of tensor decomposition." @default.
- W4229011569 created "2022-05-08" @default.
- W4229011569 creator A5003311970 @default.
- W4229011569 creator A5033106393 @default.
- W4229011569 creator A5049129395 @default.
- W4229011569 date "2022-04-25" @default.
- W4229011569 modified "2023-10-03" @default.
- W4229011569 title "PTD" @default.
- W4229011569 cites W1529395305 @default.
- W4229011569 cites W1870083876 @default.
- W4229011569 cites W1963826206 @default.
- W4229011569 cites W1989881099 @default.
- W4229011569 cites W2018282388 @default.
- W4229011569 cites W2054626033 @default.
- W4229011569 cites W2113055885 @default.
- W4229011569 cites W2136002544 @default.
- W4229011569 cites W2262232454 @default.
- W4229011569 cites W2412509443 @default.
- W4229011569 cites W2528907418 @default.
- W4229011569 cites W2547420815 @default.
- W4229011569 cites W2572561073 @default.
- W4229011569 cites W2603123944 @default.
- W4229011569 cites W2777194773 @default.
- W4229011569 cites W2780624730 @default.
- W4229011569 cites W2801848240 @default.
- W4229011569 cites W2879216670 @default.
- W4229011569 cites W2896689692 @default.
- W4229011569 cites W2906447902 @default.
- W4229011569 cites W2911424785 @default.
- W4229011569 cites W2915438736 @default.
- W4229011569 cites W2942074357 @default.
- W4229011569 cites W2943520232 @default.
- W4229011569 cites W2962858109 @default.
- W4229011569 cites W2963678090 @default.
- W4229011569 cites W2982058372 @default.
- W4229011569 cites W2997520009 @default.
- W4229011569 cites W3036644531 @default.
- W4229011569 cites W3038930935 @default.
- W4229011569 cites W3071470454 @default.
- W4229011569 cites W3086623482 @default.
- W4229011569 cites W3092879151 @default.
- W4229011569 cites W3092997524 @default.
- W4229011569 cites W3093010840 @default.
- W4229011569 cites W3093787659 @default.
- W4229011569 cites W3096651815 @default.
- W4229011569 cites W3102988878 @default.
- W4229011569 cites W3105446653 @default.
- W4229011569 cites W3125803510 @default.
- W4229011569 cites W3135133133 @default.
- W4229011569 cites W3158734383 @default.
- W4229011569 cites W3177027279 @default.
- W4229011569 cites W3179079440 @default.
- W4229011569 cites W3205798727 @default.
- W4229011569 cites W3207556837 @default.
- W4229011569 doi "https://doi.org/10.1145/3477314.3507036" @default.
- W4229011569 hasPublicationYear "2022" @default.
- W4229011569 type Work @default.
- W4229011569 citedByCount "3" @default.
- W4229011569 countsByYear W42290115692022 @default.
- W4229011569 countsByYear W42290115692023 @default.
- W4229011569 crossrefType "proceedings-article" @default.
- W4229011569 hasAuthorship W4229011569A5003311970 @default.
- W4229011569 hasAuthorship W4229011569A5033106393 @default.
- W4229011569 hasAuthorship W4229011569A5049129395 @default.
- W4229011569 hasConcept C119857082 @default.
- W4229011569 hasConcept C123201435 @default.
- W4229011569 hasConcept C124101348 @default.
- W4229011569 hasConcept C124681953 @default.
- W4229011569 hasConcept C144024400 @default.
- W4229011569 hasConcept C154945302 @default.
- W4229011569 hasConcept C155281189 @default.
- W4229011569 hasConcept C162324750 @default.
- W4229011569 hasConcept C187736073 @default.
- W4229011569 hasConcept C18903297 @default.
- W4229011569 hasConcept C202444582 @default.
- W4229011569 hasConcept C2164484 @default.
- W4229011569 hasConcept C2779304628 @default.
- W4229011569 hasConcept C2780451532 @default.
- W4229011569 hasConcept C2986737658 @default.
- W4229011569 hasConcept C33923547 @default.
- W4229011569 hasConcept C36289849 @default.
- W4229011569 hasConcept C38652104 @default.
- W4229011569 hasConcept C41008148 @default.
- W4229011569 hasConcept C42704193 @default.
- W4229011569 hasConcept C76155785 @default.
- W4229011569 hasConcept C86803240 @default.
- W4229011569 hasConceptScore W4229011569C119857082 @default.
- W4229011569 hasConceptScore W4229011569C123201435 @default.
- W4229011569 hasConceptScore W4229011569C124101348 @default.
- W4229011569 hasConceptScore W4229011569C124681953 @default.
- W4229011569 hasConceptScore W4229011569C144024400 @default.
- W4229011569 hasConceptScore W4229011569C154945302 @default.
- W4229011569 hasConceptScore W4229011569C155281189 @default.
- W4229011569 hasConceptScore W4229011569C162324750 @default.
- W4229011569 hasConceptScore W4229011569C187736073 @default.
- W4229011569 hasConceptScore W4229011569C18903297 @default.
- W4229011569 hasConceptScore W4229011569C202444582 @default.
- W4229011569 hasConceptScore W4229011569C2164484 @default.
- W4229011569 hasConceptScore W4229011569C2779304628 @default.
- W4229011569 hasConceptScore W4229011569C2780451532 @default.