Details

July 21, 2023

10:50 am

-

11:30 am

Union

Track:

Governance & Risk

Level:

Intermediate

Description

Do you want to produce AIML models that are effective in practice, and can be applied to a broader set of inputs than the models were originally trained on? Do you need access to sensitive data to train and validate your AIML models? The solution to both may surprise you.

There is a mathematical property that defines an adjustable information limit. And that information limit can be leveraged to avoid overfitting models (making those models generalizable) while also protecting data (making sensitive data accessible). It’s called differential privacy, a technical model that protects data by requiring that the information contributed by any individual does not significantly affect the output. That individual can be a person or thing that is contributing confidential information.

Building fair models that reuse sensitive data can improve services, identify new opportunities and insights that can shape an organization, and create data products that serve the needs of business and society. With differential privacy, and a framework of risk metrics and other associated benchmarks, we can enable the save and responsible reuse of data. This presentation will show you how.

Details

July 21, 2023

10:50 am

-

11:30 am

Union

Track:

Governance & Risk

Level:

Intermediate

Presenters
Devyani Biswal
Methodology Architect
Privacy Analytics (an IQVIA company)

Bio

Devyani Biswal is Methodology Architect and AI Scientist at Privacy Analytics (an IQVIA company). She supports clients in the safe and responsible uses of sensitive data across the data lifecycle. While her primary set of tools are statistical in nature, she uses any other relevant frameworks for data enablement. Devyani is also an award-winning doctoral student whose research explores the application of differential privacy and other privacy models to popular statistical learning methods. Before entering the field of privacy and data protection, she also completed research into financial risk metrics. Passionate about making complex ideas digestible to a wide audience, Devyani writes a monthly deep learning newsletter for employees. In her newsletter, she covers popular deep learning methods while exercising her amateur cartooning skills.

Bio

Devyani Biswal is Methodology Architect and AI Scientist at Privacy Analytics (an IQVIA company). She supports clients in the safe and responsible uses of sensitive data across the data lifecycle. While her primary set of tools are statistical in nature, she uses any other relevant frameworks for data enablement. Devyani is also an award-winning doctoral student whose research explores the application of differential privacy and other privacy models to popular statistical learning methods. Before entering the field of privacy and data protection, she also completed research into financial risk metrics. Passionate about making complex ideas digestible to a wide audience, Devyani writes a monthly deep learning newsletter for employees. In her newsletter, she covers popular deep learning methods while exercising her amateur cartooning skills.

Bio

Bio

Bio

Bio

Bio

Bio