Sauvik Das
Carnegie Mellon University, PA·Computer Science·Early-career
Photos that people share online can be scanned by software that recognizes faces and guesses personal details. More broadly, AI products collect data about their users and make guesses about them, and those guesses can be intrusive or simply wrong.
- Fields
Computer Science
Social Science
- Topics
usable privacy
adversarial machine learning
facial recognition
- Grants
- 2 known grants, through Sep 2027
What they work on
- Create a standardized set of measures that human evaluators use to rate the quality and acceptability of images altered by adversarial machine learning, and publish a public repository of these ratings seeded with evaluations of existing anti-surveillance tools.
- Develop a differentiable loss term, a penalty a training algorithm can optimize directly, that captures the visual appeal of altered images, and release it as open-source code for a popular machine-learning library.
- Co-design with advocacy organizations an app that lets users touch up photos they share so the images look good and also evade facial recognition and other automated inference of personal information.
- Build a taxonomy of algorithmic privacy intrusions that connects the privacy harms of consumer AI to the specific capabilities and data requirements of AI systems.
- Develop the Privacy through Design (PtD) method iteratively with industry and academic practitioners, producing a guidebook with worksheets and a repository of real design cases.
- Formally compare consumer AI products designed with PtD against products designed with existing industry standards, to test whether PtD better balances privacy concerns with usefulness.
More about the research
Two projects ask how technology can be designed so that ordinary people keep control over their privacy. One builds tools that let people protect their own images from automated surveillance. The other gives designers of consumer AI a structured way to spot and reduce privacy harms before a product is built.
Why it matters
People who post photos online, especially groups at higher risk from automated surveillance, could gain practical tools that weaken facial recognition without ruining how their images look. Companies building consumer AI could use the PtD guidebook and design cases to avoid harms such as wrong inferences about sensitive personal traits or facial recognition databases assembled without people's consent.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Cori Faklaris
University of North Carolina at Charlotte, NC·Computer Science
Gather privacy experts' opinions on how a centralized repository for tracking privacy harms, such as online harassment and cyber abuse, should be designed.
Christopher Bryan
Arizona State University Campus Immersion, AZ·Computer Science
Develop a visualization grammar, a way to specify chart designs without low-level programming, that supports privacy constraints and data sanitization actions.
Prerna Juneja
Seattle University, WA·Computer Science
Build a taxonomy of user interactions with AI companion apps, covering motivations, challenges and values, by analyzing large-scale social media data and user interviews.
The school
Carnegie Mellon University: 28 labs on PI Pages, $505M a year on research, 346 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.