|
Activity Number:
|
465
- Privacy, Confidentiality, and Disclosure Limitation
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Government Statistics Section
|
|
Abstract #314100
|
|
|
Title:
|
Differential Privacy for Model Sharing Between Government Agencies
|
|
Author(s):
|
Ellen Galantucci* and Alex Measure and David Oh
|
|
Companies:
|
BLS and Bureau of Labor Statistics and BLS
|
|
Keywords:
|
differential privacy;
neural network ;
disclosure limitation
|
|
Abstract:
|
The Survey of Occupational Injuries and Illnesses (SOII) at the Bureau of Labor Statistics (BLS) uses a neural networks autocoder to code the majority of collected injury and illness information for approximately 230,000 establishments per year. Because of the success of this method and our large training dataset, other government agencies have inquired into whether the autocoder could be made available outside of the BLS for use in other surveys and programs. Due to the nature of the training data, it may be possible to reidentify certain respondents in the training data if the autocoder were to be released as it is. To allow for the possibility of sharing the autocoder, we have implemented differential privacy on the model, which prevents reidentification and protects our respondents.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2020 program
|
