Context-Aware Generative Adversarial Privacy
Preserving the utility of published datasets while simultaneously providing provable privacy guarantees is a well-known challenge. On the one hand, context-free privacy solutions, such as differential privacy, provide strong privacy guarantees, but often lead to a significant reduction in utility. O...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2017-12-01
|
| Series: | Entropy |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1099-4300/19/12/656 |
| _version_ | 1811299959876616192 |
|---|---|
| author | Chong Huang Peter Kairouz Xiao Chen Lalitha Sankar Ram Rajagopal |
| author_facet | Chong Huang Peter Kairouz Xiao Chen Lalitha Sankar Ram Rajagopal |
| author_sort | Chong Huang |
| collection | DOAJ |
| description | Preserving the utility of published datasets while simultaneously providing provable privacy guarantees is a well-known challenge. On the one hand, context-free privacy solutions, such as differential privacy, provide strong privacy guarantees, but often lead to a significant reduction in utility. On the other hand, context-aware privacy solutions, such as information theoretic privacy, achieve an improved privacy-utility tradeoff, but assume that the data holder has access to dataset statistics. We circumvent these limitations by introducing a novel context-aware privacy framework called generative adversarial privacy (GAP). GAP leverages recent advancements in generative adversarial networks (GANs) to allow the data holder to learn privatization schemes from the dataset itself. Under GAP, learning the privacy mechanism is formulated as a constrained minimax game between two players: a privatizer that sanitizes the dataset in a way that limits the risk of inference attacks on the individuals’ private variables, and an adversary that tries to infer the private variables from the sanitized dataset. To evaluate GAP’s performance, we investigate two simple (yet canonical) statistical dataset models: (a) the binary data model; and (b) the binary Gaussian mixture model. For both models, we derive game-theoretically optimal minimax privacy mechanisms, and show that the privacy mechanisms learned from data (in a generative adversarial fashion) match the theoretically optimal ones. This demonstrates that our framework can be easily applied in practice, even in the absence of dataset statistics. |
| first_indexed | 2024-04-13T06:43:22Z |
| format | Article |
| id | doaj.art-e3462165b2e94478bdcb4a6a9c9bd4b1 |
| institution | Directory Open Access Journal |
| issn | 1099-4300 |
| language | English |
| last_indexed | 2024-04-13T06:43:22Z |
| publishDate | 2017-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Entropy |
| spelling | doaj.art-e3462165b2e94478bdcb4a6a9c9bd4b12022-12-22T02:57:40ZengMDPI AGEntropy1099-43002017-12-01191265610.3390/e19120656e19120656Context-Aware Generative Adversarial PrivacyChong Huang0Peter Kairouz1Xiao Chen2Lalitha Sankar3Ram Rajagopal4School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85281, USADepartment of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305, USADepartment of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305, USASchool of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85281, USADepartment of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305, USAPreserving the utility of published datasets while simultaneously providing provable privacy guarantees is a well-known challenge. On the one hand, context-free privacy solutions, such as differential privacy, provide strong privacy guarantees, but often lead to a significant reduction in utility. On the other hand, context-aware privacy solutions, such as information theoretic privacy, achieve an improved privacy-utility tradeoff, but assume that the data holder has access to dataset statistics. We circumvent these limitations by introducing a novel context-aware privacy framework called generative adversarial privacy (GAP). GAP leverages recent advancements in generative adversarial networks (GANs) to allow the data holder to learn privatization schemes from the dataset itself. Under GAP, learning the privacy mechanism is formulated as a constrained minimax game between two players: a privatizer that sanitizes the dataset in a way that limits the risk of inference attacks on the individuals’ private variables, and an adversary that tries to infer the private variables from the sanitized dataset. To evaluate GAP’s performance, we investigate two simple (yet canonical) statistical dataset models: (a) the binary data model; and (b) the binary Gaussian mixture model. For both models, we derive game-theoretically optimal minimax privacy mechanisms, and show that the privacy mechanisms learned from data (in a generative adversarial fashion) match the theoretically optimal ones. This demonstrates that our framework can be easily applied in practice, even in the absence of dataset statistics.https://www.mdpi.com/1099-4300/19/12/656generative adversarial privacygenerative adversarial networksprivatizer networkadversarial networkstatistical data privacydifferential privacyinformation theoretic privacymutual information privacyerror probability gamesmachine learning |
| spellingShingle | Chong Huang Peter Kairouz Xiao Chen Lalitha Sankar Ram Rajagopal Context-Aware Generative Adversarial Privacy Entropy generative adversarial privacy generative adversarial networks privatizer network adversarial network statistical data privacy differential privacy information theoretic privacy mutual information privacy error probability games machine learning |
| title | Context-Aware Generative Adversarial Privacy |
| title_full | Context-Aware Generative Adversarial Privacy |
| title_fullStr | Context-Aware Generative Adversarial Privacy |
| title_full_unstemmed | Context-Aware Generative Adversarial Privacy |
| title_short | Context-Aware Generative Adversarial Privacy |
| title_sort | context aware generative adversarial privacy |
| topic | generative adversarial privacy generative adversarial networks privatizer network adversarial network statistical data privacy differential privacy information theoretic privacy mutual information privacy error probability games machine learning |
| url | https://www.mdpi.com/1099-4300/19/12/656 |
| work_keys_str_mv | AT chonghuang contextawaregenerativeadversarialprivacy AT peterkairouz contextawaregenerativeadversarialprivacy AT xiaochen contextawaregenerativeadversarialprivacy AT lalithasankar contextawaregenerativeadversarialprivacy AT ramrajagopal contextawaregenerativeadversarialprivacy |