Resources

In this blog post, we outline the three key desiderata of synthetic data solutions — flexibility, accuracy, and privacy — and explain the fundamental trade-off between them.

Learn how the AIM algorithm, co-invented by Tumult Labs CEO Gerome Miklau, improves upon existing algorithms for synthetic data generation by adapting to the user’s analysis needs and capturing key patterns in the input data.

A scientific paper, co-authored by our CEO Gerome Miklau, introduces a cutting-edge method for generating differentially private synthetic data.

Tumult Labs helped engineers at the Wikimedia Foundation design, implement, and deploy a differentially private solution to publish Wikipedia usage metrics in a provably secure way.

Watch a teaser demo of Tumult Tune, an upcoming product that allows you to easily understand and optimize differentially private data products.

In this blog post, we introduce a conceptual framework to help prospective buyers of anonymization technology evaluate claims on the trustworthiness and security of potential solutions.

In a talk for PEPR ‘24, Damien Desfontaines lists major issues with empirical privacy metrics for synthetic data generation, and explains how we could fix them.

In a talk for PEPR ‘24, Daniel Simmons-Marengo explains why anonymization is at risk, and what we can do to safeguard user trust going forward.

Researchers at University of Vermont ran a usability study to compare various differential privacy tools. Can you guess which platform study participants found easiest to use correctly?

Read the technical paper describing Tumult Analytics’ design goals and its architecture.

When the U.S. Census Bureau needed to release earnings data post-secondary graduates, Tumult Labs Chief Scientist and co-founder Ashwin Machanavajjhala helped the agency design a new privacy-safe algorithm.

We are proud to be named a strategic partner in Google’s latest initiative to make BigQuery data clean rooms available in public preview, announced last week at NEXT ‘23.

Damien Desfontaines describes how differential privacy can bring together open data with safe, privacy-preserving publication practices.

In this blog post, we review legacy techniques used to anonymize data. We give real-world examples of the failure modes of these techniques. Then, we draw some lessons from these historical failures.

Tumult Labs was named a Sample Vendor in differential privacy across all three reports: Privacy, Data Security, and Data Science & Machine Learning.

In this article, we first explain what it means for differential privacy software to have vulnerabilities. Then, we present a new class of attacks on naive implementations of differential privacy: precision-based attacks. These attacks expose vulnerabilities in multiple open-source libraries, including diffprivlib, SmartNoise Core, and OpenDP.

Tumult Analytics, a Python library making it easy and safe to use differential privacy, is now open-source!

For the first time, anyone interested in better understanding Wikipedia page view data at a more granular, country-specific level can find regularly published data and useful insights.

While differential privacy offers robust privacy protection, it can sometimes unfairly impact certain groups. Read the research co-authored by the Tumult Labs founders on how to address this problem.

Say you’re working on a new project involving sensitive data — for example, adding a new feature to a healthcare app.

Tumult Analytics got a big upgrade: support for privacy IDs, allowing users to build differentially private algorithms that provide user-level privacy guarantees, even when each user may contribute multiple records.

Read the paper co-authored by Tumult Labs founders on building a differentially private relational database system that takes into account the complexity of multi-relational schemas and constraints.

Using data clean rooms to join data between two parties does not always mean that the data is fully protected: outputs can also sometimes leak individual information. How can you mitigate this risk?

Some data publication or sharing use cases are well-suited to the use of differential privacy, while some aren’t. Learn a quick litmus test allowing you to quickly distinguish between the two.

We are teaming up with Google Cloud to bring differential privacy to BigQuery and the Google Cloud Platform. Learn more and get started in just a few minutes!

How to publish the output of large workloads of queries on datasets with many dimensions, achieving accuracy and scalability with strong privacy guarantees? Read the research co-authored by Tumult Labs founders

How to evaluate the disparate impact of privacy-preserving data analytics on different population groups? This research paper by Tumult Labs' Chief Scientist Ashwin Machanavajjhala explores this question.

Designing a programming framework for writing complex yet safe differential privacy programs is no small task. This paper co-authored by Tumult Labs founders laid the foundation of the privacy framework used by our customers.

Tumult Labs designed a novel differentially-private algorithm that the U.S. Census Bureau is using to publish the Detailed Demographic and Housing Characteristics (DHC) Race & Ethnicity tabulations, as part of the 2020 Census.

Read the research co-authored by Tumult Labs co-founder Ashwin Machanavajjhala on how to formalize complex privacy requirements mandated by law using novel notions and algorithms based on differential privacy.

Which synthetic data generation algorithms for tabular datasets offer the best privacy/utility trade-offs? Tumult Labs did the research. Read the results below.

Watch a short demo video outlining the main features of Tumult Analytics, and demonstrating the ease-of-use of its interface.