One year of Hiding Nemo
Taking stock after one year doing independent consulting on privacy-enhancing technology.
Ted is writing things
Latest posts
Banning noise will be a disaster for statistical data products
Sadly, taking away valuable disclosure avoidance tools doesn't make fundamental trade-offs go away.
Guest post: Goodbye (ε,δ), hello μ! Reporting privacy guarantees in machine learning
Why is Gaussian DP a much better compact representation of privacy guarantees than (ε,δ)-DP, especially for ML applications.
Notes from Dagstuhl: biomedical data sharing
I learned a ton about data sharing and privacy issues in the biomedical field, here are some of my takeaways.
Research highlight: Cliopatra: Extracting Private Information from LLM Insights
The price of not using robust notions when building “privacy-preserving” analytics systems.
Research post: Privacy in theory, bugs in practice
Findings a bunch of bugs in differential privacy libraries using a new approach to privacy auditing.
Guest post: Balancing your dataset? Mind the privacy leaks!
Why SMOTE, a popular data augmentation technique, exposes your sensitive data.
Three kinds of “privacy auditing”
What academics call “privacy auditing” has little to do with what industry folks call “auditing”. This may be confusing.
On “simplifying” GDPR's definition of personal data
I don't think it's a great idea to redefine what counts as anonymized data according to GDPR, like the European Commission is currently proposing. Here's why.
Favorite reads of 2025
A list of the books that I particularly enjoyed reading this year.
Mapping ISO standards to modern privacy attacks
ISO standards on privacy attacks are not great, but some organizations still want to use them. Can we interpret them in a principled way?
Tell me what to write
A short note encouraging you to tell me what you would like to read about in this blog.
Research highlights: Privacy attacks on statistical data
A summary of past and recent work on privacy attacks against statistical data releases.
Five hard lessons learned by privacy engineers
Turns out, AI isn't that special.
Answering the BfDI's questions on personal data in LLMs
The German data protection authority asked me for my input on technical questions regarding the use of personal data in AI models; here are my answers.
Introducing Hiding Nemo 🪸
A consultancy focused on helping organizations do more with data while deeply respecting their users, using privacy-enhancing technology.
Can you beat me at WORDPL?
What if Wordle, but with randomized response? Let's learn about Bayesian inference to get on top of the leaderboard, ponder what an optimal strategy might be… and maybe win some chocolate?
The search for better empirical privacy metrics
A few generic pieces of advice on how to get better utility out of your differentially private aggregations.
A glossary of differential privacy terms
A list of short definitions of commonly used terms in differential privacy, with references for further reading.
Don't use diffprivlib
A critical examination of an open-source differential privacy library.
Five things privacy experts know about AI
… and that AI salespeople don't want you to know!
Winning at DP VISION — suspiciously easily…
I crafted a strategy to beat the high score at a game based on differential privacy. It should have been harder to beat, though…
Paper highlights: Utility-boosting geometric tricks
A quick look at three papers who use neat tricks to boost the utility of simple DP operations.
What's up with all these large privacy budgets?
Many real-world DP deployments use privacy parameters that can seem unconvincing. Should we be worried?
Empirical privacy metrics: the bad, the ugly… and the good, maybe?
This post contains the slides and transcript of a talk about empirical privacy metrics that I delivered at PEPR in June 2024.
Converters between differential privacy variants
A small collection of interactive converters between differential privacy variants.
Paper highlight: Evaluations of Machine Learning Privacy Defenses are Misleading
A quick look at a new paper poking at empirical privacy metrics for ML models.
Five stages of accepting provably robust anonymization
This post contains the slides and transcript for an invited talk I delivered at AnoSiDat in April 2024.
Choosing things privately with the exponential mechanism
A simple introduction to an essential building block for differential privacy: how to select a value among many.
Mapping privacy-enhancing technologies to your use cases
A guide listing common privacy-enhancing technologies, and delineating between which problem each one solves.
What anonymization techniques can you trust?
An overview of legacy techniques used to anonymize data, how they fail, and what we can learn from these failures.
Is differential privacy the right fit for your problem?
Some data publication or sharing use cases are well-suited to the use of differential privacy, while some aren’t. In this blog post, we give a litmus test allowing you to quickly distinguish between the two.
Research post: Differential privacy under partial knowledge
What happens to differential privacy if you model a weaker adversary with only partial knowledge over the input data?
A bottom-up approach to making differential privacy ubiquitous
This post contains the slides and speaker notes for an invited talk I delivered at PPAI-22.
Averaging risk: Rényi DP & zero-concentrated DP
Averaging the privacy loss random variable across outputs can be useful: introducing Rényi DP, and zero-concentrated DP.
A list of real-world uses of differential privacy
A list of practical deployments of differential privacy, along with their privacy parameters.
A friendly, non-technical introduction to differential privacy
An introduction and table of contents for my beginner-friendly blog post series about differential privacy.
Joining Tumult Labs
I'm joining Tumult Labs, a startup focused on differential privacy. Here's why I'm excited about it!
Don't worry, your data's noisy
Your data was already noisy before I got a chance to add noise to it! Here's why you shouldn't panic, and also what you should do about it.
Getting more useful results with differential privacy
A few generic pieces of advice on how to get better utility out of your differentially private aggregations.
Demystifying the US Census Bureau's reconstruction attack
The US Census is moving to differential privacy, after running a successful privacy attack on their 2010 release. Let's look at this attack in more detail!
Why not differential privacy?
What does it mean for an algorithm to not be differentially private?
Converting my PhD thesis into HTML
A story of false hopes, perseverance, pain, and futility.
The magic of Gaussian noise
Why is Gaussian noise a popular choice to make statistics and machine learning models differentially private?
The privacy loss random variable
What does \(\delta\) really mean in \((\varepsilon,\delta)\) -differential privacy? Let's explain this using a central concept: the privacy loss random variable.
A reading list on differential privacy
A list of papers to check out to learn more about differential privacy.
« What does a privacy engineer do, anyway? »
Besides having a super cool job title, what is it like being a privacy engineer?
Local vs. central differential privacy
Differential privacy is used in two very distinct contexts. Come learn about the distinction between the two, and interesting new directions that combine them!
Research post: Cardinality Estimators do not Preserve Privacy
You can't both remember unique individuals and not remember them. Shocking, right? :D
Almost differential privacy
Publishing histograms without knowing the categories in advance: introducing (ε,δ)-differential privacy.