IP Library › Granted Patent US 12,493,718
Granted Patent B2
US 12,493,718 · App. 18/324,484 · Granted Dec 9, 2025

Providing data collaboration via a distributed secure collaboration framework

Inventors: Sunav Choudhary (Kolkata, IN); Subrata Mitra (Bangalore, IN); Sanjay Sukumaran (Kanhangad, IN); Priyanshu Yadav (Jaipur, IN); Munish Gupta (Shahdol, IN); Jashn Arora (Faridabad, IN); Iftikhar Ahamath Burhanuddin (Bangalore, IN); Gautam Choudhary (Sri Ganganagar, IN); Atharv Tyagi (New Delhi, IN)
Assignee: Adobe Inc.
G06F21/6254
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,493,718
App. No.
18/324,484
Granted
Dec 9, 2025
Kind
B2
Abstract

The present disclosure relates to systems, methods, and non-transitory computer-readable media that implements a secure distributed data collaboration architecture for generating synthetic datasets. For example, the disclosed system sends a request to perform a data collaboration with a first dataset of a first local node and a second dataset of a second local node. The disclosed system receives intermediate feature maps from the local nodes that correspond with the datasets and generates a combined feature map. Further, the disclosed system generates a synthetic dataset from the combined feature map by utilizing a central generative model. Moreover, the synthetic dataset generated by the disclosed system is statistically representative of the first dataset and the second dataset.

Claims (74)

1 . A computer-implemented method comprising:

sending a request to perform a data collaboration with a first dataset from a first local node and a second dataset from a second local node, wherein the first dataset and the second dataset comprises personally identifiable information;

performing pre-processing of the first dataset from the first local node and the second dataset from the second local node by utilizing a private set intersection model to determine an overlap of users without exposing raw information of the first dataset to the second local node and without exposing raw information of the second dataset to the first local node;

receiving a first intermediate feature map corresponding with the first dataset from the first local node without personally identifiable information, wherein the first intermediate feature map is generated by transforming, via a first local generator, one or more columns of the first dataset and the first intermediate feature map statistically represents the first dataset;

receiving a second intermediate feature map corresponding with the second datasets from the second local node without personally identifiable information,

wherein the second intermediate feature map is generated by transforming, via a second local generator, one or more columns of the second dataset and the second intermediate feature map statistically represents the second dataset,

wherein personally identifiable information from the first local node stays siloed at the first local node and personally identifiable information from the second local node stays siloed at the second local node while still generating the first intermediate feature map that statistically represents the first dataset and generating the second intermediate feature map that statistically represents the second dataset;

generating a combined feature map from the first intermediate feature map and the second intermediate feature map; and

generating, utilizing a central generative model, a synthetic dataset from the combined feature map, wherein the synthetic dataset is statistically representative of the first dataset and the second dataset.

2 . The computer-implemented method of claim 1 , wherein generating the synthetic dataset comprises:

utilizing the central generative model as a centralized server to receive the combined feature map, wherein the first local node and the second local node are remote distributed devices relative to the central generative model; and

generating, utilizing the central generative model, the synthetic dataset, wherein personally identifiable information from the first local node and the second local node is not exposed to the central generative model.

3 . The computer-implemented method of claim 2 , further comprising:

training the central generative model, the first local generator, and the second local generator by:

determining measures of loss for the first local generator, the second local generator, and the central generative model; and

modifying parameters of the first local generator, the second local generator, and the central generative model based on the determined measures of loss.

4 . The computer-implemented method of claim 3 , wherein training the central generative model, the first local generator, and the second local generator further comprises utilizing conditional vector sampling to account for datasets with skewed category frequencies.

5 . The computer-implemented method of claim 1 , wherein generating the combined feature map comprises utilizing the central generative model to combine the first intermediate feature map and the second intermediate feature map.

6 . The computer-implemented method of claim 1 , further comprising:

transforming, utilizing a transformer, discrete columns of the first dataset and discrete columns from the second dataset to columns corresponding to a number of categories from the discrete columns of the first dataset and a number of categories of the discrete columns from the second dataset; and

transforming, utilizing the transformer, continuous columns of the first dataset and continuous columns of the second dataset to an approximate value column.

7 . The computer-implemented method of claim 1 , further comprising generating the combined feature map by utilizing a mixing matrix to mix the first intermediate feature map and the second intermediate feature map.

8 . The computer-implemented method of claim 1 , further comprising determining a correlation between various rows of the first intermediate feature map and the second intermediate feature map to generate the synthetic dataset.

9 . A system comprising:

one or more memory devices; and

one or more processors configured to cause the system to:

receive, from a client device, a request to perform a data collaboration between a first dataset from the client device and a second dataset from a local node comprising personally identifiable information;

perform pre-processing of the first dataset from the client device and the second dataset from the local node by utilizing a private set intersection model to determine an overlap of users without exposing raw information of the first dataset to the local node and without exposing raw information of the second dataset to the client device;

generate, via a generator of the client device, a first intermediate feature map without personally identifiable information, wherein the first intermediate feature map is generated by transforming, via a first local generator, one or more columns of the first dataset and the first intermediate feature map statistically represents the first dataset;

generate, via a generator of the local node, a second intermediate feature map without personally identifiable information,

wherein the second intermediate feature map is generated by transforming, via a second local generator, one or more columns of the second dataset and the second intermediate feature map statistically represents the second dataset,

wherein personally identifiable information from the client device stays siloed at the client device and personally identifiable information from the local node stays siloed at the local node while still generating the first intermediate feature map that statistically represents the first dataset and generating the second intermediate feature map that statistically represents the second dataset;

generate a combined feature map from the first intermediate feature map and the second intermediate feature map;

generate, utilizing a central generative model, a synthetic dataset from the combined feature map, the synthetic dataset comprising a statistically representative dataset of the first dataset and the second dataset; and

provide the synthetic dataset to the client device.

10 . The system of claim 9 , wherein the one or more processors are configured to cause the system to generate the combined feature map using the central generative model by combining the first intermediate feature map and the second intermediate feature map and the central generative model is a centralized server.

11 . The system of claim 9 , wherein the one or more processors are configured to cause the system to generate, utilizing the central generative model, the synthetic dataset, wherein personally identifiable information from the client device and the local node is not exposed to the central generative model.

12 . The system of claim 9 , wherein the one or more processors are configured to cause the system to utilize a transformer to:

transform a first discrete column of the first dataset to columns corresponding to a number of categories of the first discrete column of the first dataset; and

transform a first discrete column of the second dataset to columns corresponding to a number of categories of the first discrete column of the second dataset.

13 . The system of claim 9 , wherein the one or more processors are configured to cause the system to:

utilize a transformer to transform a first continuous column of the first dataset and a first continuous column of the second dataset to an approximate value column by:

determining a difference between a first probability distribution statistic and each value of the first continuous column of the first dataset; and

determining a difference between a second probability distribution statistic and each value of the first continuous column of the second dataset.

14 . The system of claim 9 , wherein the one or more processors are configured to cause the system to:

generate the combined feature map by utilizing a mixing matrix to mix the first intermediate feature map and the second intermediate feature map; and

utilize the central generative model to generate the synthetic dataset by determining a correlation between various rows of the first intermediate feature map and various rows of the second intermediate feature map.

15 . A non-transitory computer-readable medium storing executable instructions which, when executed by a processing device, cause the processing device to perform operations comprising:

performing pre-processing of a first dataset from a first local node and a second dataset from a second local node by utilizing a private set intersection model to determine an overlap of users without exposing raw information of the first dataset to the second local node and without exposing raw information of the second dataset to the first local node;

receiving a first intermediate feature map generated from the first dataset from the first local node without personally identifiable information by transforming, via a first local generator, one or more columns of the first dataset and the first intermediate feature map statistically represents the first dataset;

receiving a second intermediate feature map generated from the second dataset from the second local node without personally identifiable information by transforming, via a second local generator, one or more columns of the second dataset and the second intermediate feature map statistically represents the second dataset,

wherein personally identifiable information from the first local node stays siloed at the first local node and personally identifiable information from the second local node stays siloed at the second local node while still generating the first intermediate feature map that statistically represents the first dataset and generating the second intermediate feature map that statistically represents the second dataset;

generating a combined feature map from the first intermediate feature map and the second intermediate feature map by utilizing a mixing matrix to mix the first intermediate feature map and the second intermediate feature map; and

generating, utilizing a central generative model, a synthetic dataset from the combined feature map by determining a correlation between various rows of the first intermediate feature map and the second intermediate feature map, wherein the synthetic dataset is statistically representative of the first dataset and the second dataset.

16 . The non-transitory computer-readable medium of claim 15 , wherein:

generating the combined feature map comprises utilizing the central generative model to combine the first intermediate feature map and the second intermediate feature map; and

generating the synthetic dataset comprises:

utilizing the central generative model as a centralized server to receive the combined feature map, wherein the first local node and the second local node are remote distributed devices relative to the central generative model; and

generating, utilizing the central generative model, the synthetic dataset, wherein personally identifiable information from the first local node and the second local node is not exposed to the central generative model.

17 . The non-transitory computer-readable medium of claim 15 , wherein the operations further comprise:

transforming a first continuous column of the first dataset to an approximate value column by determining a difference between each value of the first continuous column of the first dataset and a first probability distribution statistic; and

transforming a first continuous column of the second dataset to the approximate value column by determining a difference between each value of the first continuous column of the second dataset and a second probability distribution statistic.

18 . The non-transitory computer-readable medium of claim 15 , wherein the operations further comprise:

utilizing a transformer to transform a first discrete column of the first dataset to columns corresponding to a number of categories of the first discrete column of the first dataset; and

utilizing the transformer to transform a first discrete column of the second dataset to columns corresponding to a number of categories of the first discrete column of the second dataset.

19 . The non-transitory computer-readable medium of claim 15 , wherein the operations further comprise:

determining a first discriminator loss for a first local discriminator of the first local node; and

determining a second discriminator loss for a second local discriminator of the second local node.

20 . The non-transitory computer-readable medium of claim 15 , wherein the operations further comprise:

updating parameters of the central generative model by:

determining a first local generator loss to update parameters of the first local generator;

determining a second local generator loss to update parameters of the second local generator;

determining a combined measure of loss based on the first local generator loss, the second local generator loss and the synthetic dataset; and

back-propagating the combined measure of loss to the central generative model.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded May 26, 2023
From: CHOUDHARY, SUNAV; MITRA, SUBRATA; SUKUMARAN, SANJAY; YADAV, PRIYANSHU; GUPTA, MUNISH; ARORA, JASHN; BURHANUDDIN, IFTIKHAR AHAMATH; CHOUDHARY, GAUTAM; TYAGI, ATHARV
To: ADOBE INC.
Reel/Frame 063777/0280 →
Continuity (1)
Related Publication 20240394407A1 · Nov 28, 2024
References Cited (27)
US 11797701B1 · Angelo · 2023 [cited by examiner]
US 11805167B2 · Chu · 2023 [cited by examiner]
US 11886617B1 · Du · 2024 [cited by examiner]
US 20180165475A1 · Veeramachaneni · 2018 [cited by examiner]
US 20200311294A1 · Sim-Tang · 2020 [cited by examiner]
US 20200311296A1 · Kim · 2020 [cited by examiner]
US 20210200731A1 · Aleksandrovich · 2021 [cited by examiner]
US 20210279366A1 · Choudhury · 2021 [cited by examiner]
US 20220391847A1 · Barrett · 2022 [cited by examiner]
US 20230153448A1 · Mitra · 2023 [cited by examiner]
Arjovsky, M.; Chintala, S.; and Bottou, L. 2017. Wasserstein Generative Adversarial Networks. In Precup, D.; and Teh, Y. W., eds., Proceedings of the 34th International Conference on Machine Learning, vol. 70 of Proceed… [cited by applicant]
Bay, A.; Erkin, Z.; Hoepman, J.-H.; Samardjiska, S.; and Vos, J. 2022. Practical Multi-Party Private Set Intersection Protocols. vol. 17, 1-15. doi:10.1109/TIFS.2021. 3118879. [cited by applicant]
Ceballos, I.; Sharma, V.; Mugica, E.; Singh, A.; Roman, A.; Vepakomma, P.; and Raskar, R. 2020. SplitNN-driven Vertical Partitioning doi:10.48550/ARXIV.2008.04137. URL https://arxiv.org/abs/2008.04137. [cited by applicant]
Creswell, A.; White, T.; Dumoulin, V.; Arulkumaran, K.; Sengupta, B.; and Bharath, A. A. 2018. Generative Adversarial Networks: An Overview. IEEE Signal Processing Magazine 35(1): 53-65. doi:10.1109/MSP.2017.2765202. [cited by applicant]
Dwork, C. 2008. Differential Privacy: A Survey of Results. In Agrawal, M.; Du, D.; Duan, Z.; and Li, A., eds., Theory and Applications of Models of Computation, 1-19. Berlin, Heidelberg: Springer Berlin Heidelberg. ISBN… [cited by applicant]
Hao Chen, Kim Laine, P. R. 2017. Fast Private Set Intersection from Homomorphic Encryption. 1243. [cited by applicant]
Kingma, D. P.; and Welling, M. 2014. Auto-Encoding Variational Bayes. [cited by applicant]
Lepoint, T.; Patel, S.; Raykova, M.; Seth, K.; and Trieu, N. 2021. Private Join and Compute from PIR with Default. 605. [cited by applicant]
Li, L.; Fan, Y.; Tse, M.; and Lin, K.-Y. 2020. A review of applications in federated learning. volume 149, 106854. ISSN 0360-8352. doi: https://doi.org/10.1016/j.cie. 2020.106854. URL https://www.sciencedirect.com/scien… [cited by applicant]
Lin, Z.; Khetan, A.; Fanti, G.; and Oh, S. 2018. PacGAN: The power of two samples in generative adversarial networks. In Bengio, S.; Wallach, H.; Larochelle, H.; Grauman, K.; Cesa-Bianchi, N.; and Garnett, R., eds., Adv… [cited by applicant]
Liyue Fan., H. J. 2015. A Practical Framework for Privacy-Preserving Data Analytics. In WWW '15: Proceedings of the 24th International Conference on World Wide Web, 311-321. doi: https://doi.org/10.1145/2736277.2741122. [cited by applicant]
Park, N.; Mohammadi, M.; Gorde, K.; Jajodia, S.; Park, H.; and Kim, Y. 2018. Data synthesis based on generative adversarial networks. Proceedings of the VLDB Endowment 11(10): 1071-1083. doi:10.14778/3231751.3231757. UR… [cited by applicant]
Polato, M. 2021. Federated Variational Autoencoder for Collaborative Filtering. In 2021 International Joint Conference on Neural Networks (IJCNN), 1-8. doi:10.1109/ IJCNN52387.2021.9533358. [cited by applicant]
Shokri, R.; Stronati, M.; Song, C.; and Shmatikov, V. 2016. Membership Inference Attacks against Machine Learning Models. arXiv. doi:10.48550/ARXIV.1610.05820. URL https://arxiv.org/abs/1610.05820. [cited by applicant]
Wu, Y.; Cai, S.; Xiao, X.; Chen, G.; and Ooi, B. C. 2020. Privacy preserving vertical federated learning for tree-based models. Proceedings of the VLDB Endowment 13(12): 2090-2103. doi:10.14778/3407790.3407811. URL http… [cited by applicant]
Xu, L.; Skoularidou, M.; Cuesta-Infante, A.; and Veeramachaneni, K. 2019. Modeling Tabular data using Conditional GAN. doi:10.48550/ARXIV.1907.00503. URL https://arxiv.org/abs/1907.00503. [cited by applicant]
Zhao, Z.; Kunar, A.; der Scheer, H. V.; Birke, R.; and Chen, L. Y. 2021. CTAB-GAN: Effective Table Data Synthesizing. [cited by applicant]