I’m writing a general-interest book about #DifferentialPrivacy, with attention to its history, limitations, future, and use today by government and business.

If you have suggestions for things I should mention, please let me know! https://forms.gle/pWzCGm5a1iAgbnAS9

How would differential privacy affect the accuracy of county-to-county migration estimates from the 2020 Census? Find out:

https://link.springer.com/article/10.1007%2Fs11113-021-09664-5

#NICHDImpact #DifferentialPrivacy #Privacy #Data #Census #Migration

Last was an incredible talk by Alessandro Acquisti on #privacy and the #data economy at @princetoncitp. This talk covers a number of crucial issues, but the most important one for me was #DifferentialPrivacy. Acquisti does a thorough analysis of the social cost of implementing DP in the US census, showing that it's likely relatively low and can be mitigated through policy interventions. This kind of analysis should be employed every time DP is used. Highly recommend https://www.youtube.com/watch?v=MQ3TGmWXlLU (6/6)

Next was a fabulous talk by Rachel Cummings on communicating #DifferentialPrivacy guarantees to users at the C3.ai Digital Transformation Institute. There's so much theoretical work on DP and so little on this aspect of the approach, and while I always want more on the #bias tradeoffs with this approach I still highly recommend the talk https://www.youtube.com/watch?v=PuzMKkm60dg (5/7) #privacy

Next was a short talk by Claire McKay Bowen on providing different differentially private stats and analyses for tax #data. This was a wonderful breakdown of how to thoughtfully release meaningful, sensitive data to researchers. Highly recommend https://www.youtube.com/watch?v=ENvvF6jY8f4 (8/12) #statistics #DifferentialPrivacy #privacy

Next was an intriguing talk by Ashwin Machanavajjhala on #DifferentialPrivacy #algorithms for the #census. Differential #privacy provides some nice individual guarantees, but significantly complicates group level analyses, especially for small (ready: minority) groups. I appreciated the transparency here, but it didn't make me feel any better about the real harms that differential privacy will create for marginalized groups by making them look more like the majority https://www.youtube.com/watch?v=onCyEk_-yK8 (7/9)

Analysis of 2020 census differential privacy suggests net migration estimates for 5yr age groups may only be accurate for ~1/2 of counties.
Accuracy lowest for oldest ages.

https://link.springer.com/article/10.1007%2Fs11113-021-09664-5

#Census #DifferentialPrivacy #2020 #2020Census #Age #Migration #NICHDImpact

Next was an intriguing talk by Hyunseung Kang on randomized experiments w/ #DifferentialPrivacy at #UW. I'm usually not a fan of this method b/c it can minimize minority data points, but the use case described here was compelling - extremely sensitive data that can't be disclosed or analyzed, it can be a useful tool. It's challenging to square w/traditional #stats, and this talk examines these effects to enable researchers to draw conclusions. Highly recommend https://www.youtube.com/watch?v=XBPsny4cXuM (6/8)

Does anyone know of any statistics describing how quickly/widely #differentialprivacy is being adopted?

Looking for figures that complement @tedted's excellent list of real-world uses (https://desfontain.es/privacy/real-world-differential-privacy.html)

Technical researchers sometimes (like in this paper) seem to treat privacy as a function of how readily personal data can be inferred - leading to assertions that techniques that reconstruct personal data from lower-resolution sources are better for privacy. Total privacy risk models like that used in #differentialprivacy seem to me like more honest technical solutions, even if they can be a major hassle to implement. Or is there a better (technical) way?

Paper: https://arxiv.org/abs/2301.00250

Azure Confidential Computing and Sarus Smart Privacy Solution are used to combine data from multiple parties and track financial crime, while protecting the security and privacy of personal data. Differential Privacy is implemented in data manipulation to minimize leakage risk and enable advanced detection approaches to detect criminal activities in financial transactions. https://techcommunity.microsoft.com/t5/ai-customer-engineering-team/tracking-financial-crime-with-azure-confidential-computing-and/ba-p/3711957 #AzureConfidentialComputing #SarusSmartPrivacy #DifferentialPrivacy

Next was a nice panel on the challenges of #data sharing at the #AlanTuringInstitute with Tim Watson, Fernando Lucini, Cynthia Dwork, Rachel Dugdale, and Zoe Kourtzi. Personal data sharing with any organization is fraught with ethical concerns, and even techniques like #DifferentialPrivacy can cause more problems than they solve. Lots of these topics are discussed here https://www.youtube.com/watch?v=17lYBh4dV3o (3/5)

Last week's announcement that the US Census Bureau is pausing plans to deploy differential privacy for the American Community Survey does not mark the beginning of the end for #DifferentialPrivacy — in fact, it's just the reverse.

https://www.linkedin.com/feed/update/urn:li:activity:7011741459896778752/

Census just published an update on disclosure avoidance for the American Community Survey. It's a pretty big change from where we were this time 2 years ago. "Our current assessment is that the science does not yet exist to implement a formally private solution for the ACS," vs. very ambitious timelines back in 2020. They now say differential privacy may not even be the eventual method they use. #ACS #census #differentialprivacy

Next was an all-star panel on data sharing, privacy, and #regulation at the Stigler Center for the Study of the Economy and the State with Filippo Lancieri, @Julia, John de Figueiredo, Laura Edelson, & @lior. This panel goes from #data collection ethics to #privacy protection approaches (though not harsh enough on the discriminatory effects of #DifferentialPrivacy IMO) to the current state of the law in the US and Europe, and more. Highly recommend https://www.youtube.com/watch?v=j-Pz7t5V_94 (3/5)

Last was an intriguing talk by Xuechen Li on differentially private #DeepLearning. Longtime readers would know I'm extremely skeptical of #DifferentialPrivacy, and while Li shows improvements on some well-known issues in this space I'm still left wanting. The fundamental issue is that these methods disproportionately discount minority groups, and further guaranteeing individual anonymity doesn't fix that problem https://www.youtube.com/watch?v=jFWtv81F7M8 (10/10)

A collection of useful-yet-hard-to-find results and lemmas from the #differentialprivacy literature, compiled and summarized (with proofs and discussion) by Sophia Huang, currently undergrad at @UNSWCOMPUTING@twitter.com: https://arxiv.org/abs/2211.11189

Comments, feedback, and suggestions welcome!

Do we think synthetic data is going to be the actual direction #ml #ai goes? Are global regs going to push folks towards embracing #differentialprivacy ?

Hello, fediverse!

We publish the academic open access journal Proceedings on Privacy Enhancing Technologies and host the annual Privacy Enhancing Technologies Symposium. We are a top research venue dedicated to novel applied and theoretical research into the design, analysis, experimentation, and fielding of privacy-enhancing technologies.

This account will mostly be for announcements about deadlines and events, but we'll also boost posts from our authors about their research, so follow along to learn about cool privacy things.

Please consider boosting so your pals can find us, too!

#privacy #tech #privacyTech #surveillance #anonymousCommunication #differentialPrivacy #censorshipResistance #dataProtection #HCI #trafficAnalysis #MLPrivacy #genomicPrivacy
#Introductions #TwitterMigration

Hello, fediverse!

We publish the academic open access journal Proceedings on Privacy Enhancing Technologies and host the annual Privacy Enhancing Technologies Symposium. We are a top research venue dedicated to novel applied and theoretical research into the design, analysis, experimentation, and fielding of privacy-enhancing technologies.

This account will mostly be for announcements about deadlines and events, but we'll also boost posts from our authors about their research, so follow along to learn about cool privacy things.

Please consider boosting so your pals can find us, too!

#privacy #tech #privacyTech #surveillance #anonymousCommunication #differentialPrivacy #censorshipResistance #dataProtection #HCI #trafficAnalysis #MLPrivacy #genomicPrivacy
#Introductions #TwitterMigration