This paper presents a secure aggregation system Armadillo that has disruptive resistance against adversarial clients, such that any coalition of malicious clients can affect the aggregation result only by misreporting their private inputs in a pre-defined legitimate range. Armadillo is designed for federated learning setting, where a single powerful server interacts with many weak clients iteratively to train models on client's private data. While a few prior works consider disruption resistance under such setting, for an aggregation on n clients they either require high cost per client (Chowdhury et al. CCS '22) or concretely many rounds that is logarithmic in n (Bell et al. USENIX Security '23). Although disruption resistance can be achieved generically with zero-knowledge proof techniques (which we also use in this paper), we realize an efficient system with two new designs: 1) a simple two-layer secure aggregation protocol that requires only simple arithmetic computation; 2) an agreement protocol that removes the effect of malicious clients from the aggregation with low round complexity. With these techniques, Armadillo runs in 3 rounds per aggregation (our round complexity is independent of n) with computationally lightweight server and clients.

We revisit the problem of concretely efficient secure computation of sorting and selection (e.g., maximum, median, or top-k) on secret-shared data, focusing on the case of security against a single semi-honest party. Previous solutions either have a high communication overhead or many rounds of interaction, even when allowing input-independent preprocessing.

The shuffle model has recently emerged as a popular setting for differential privacy, where clients can communicate with a central server using anonymous channels or an intermediate message shuffler. This model was also explored in the context of cryptographic tasks such as secure aggregation and private information retrieval (PIR). However, this study was almost entirely restricted to the stringent notion of information-theoretic security.

This paper introduces Flamingo, a system for secure aggregation of data across a large set of clients. In secure aggregation, a server sums up the private inputs of clients and obtains the result without learning anything about the individual inputs beyond what is implied by the final sum. Flamingo focuses on the multi-round setting found in federated learning in which many consecutive summations (averages) of model weights are performed to derive a good model. Previous protocols, such as Bell et al. (CCS ’20), have been designed for a single round and are adapted to the federated learning setting by repeating the protocol multiple times. Flamingo eliminates the need for the per-round setup of previous protocols, and has a new lightweight dropout resilience protocol to ensure that if clients leave in the middle of a sum the server can still obtain a meaningful result. Furthermore, Flamingo introduces a new way to locally choose the so-called client neighborhood introduced by Bell et al. These techniques help Flamingo reduce the number of interactions between clients and the server, resulting in a significant reduction in the end-to-end runtime for a full training session over prior work.We implement and evaluate Flamingo and show that it can securely train a neural network on the (Extended) MNIST and CIFAR-100 datasets, and the model converges without a loss in accuracy, compared to a non-private federated learning system.

This paper introduces Ibex, an advertising system that reduces the amount of data that is collected on users while still allowing advertisers to bid on real-time ad auctions and measure the effectiveness of their ad campaigns. Specifically, Ibex addresses an issue in recent proposals such as Google's Privacy Sandbox Topics API in which browsers send information about topics that are of interest to a user to advertisers and demand-side platforms (DSPs). DSPs use this information to (1) determine how much to bid on the auction for a user who is interested in particular topics, and (2) measure how well their ad campaign does for a given audience (i.e., measure conversions). While Topics and related proposals reduce the amount of user information that is exposed, they still reveal user preferences. In Ibex, browsers send user information in an encrypted form that still allows DSPs and advertisers to measure conversions, compute aggregate statistics such as histograms about users and their interests, and obliviously bid on auctions without learning for whom they are bidding. Our implementation of Ibex shows that creating histograms is 1.-2.5× more expensive for browsers than disclosing user information, and Ibex's oblivious bidding protocol can finish auctions within 550 ms. We think this makes Ibex capable of preserving a good experience while improving user privacy.

This paper introduces Mycelium, the first system to process differentially private queries over large graphs that are distributed across millions of user devices. Such graphs occur, for instance, when tracking the spread of diseases or malware. Today, the only practical way to query such graphs is to upload them to a central aggregator, which requires a great deal of trust from users and rules out certain types of studies entirely. With Mycelium, users' private data never leaves their personal devices unencrypted, and each user receives strong privacy guarantees. Mycelium does require the help of a central aggregator with access to a data center, but the aggregator merely facilitates the computation by providing bandwidth and computation power; it never learns the topology of the graph or the underlying data. Mycelium accomplishes this with a combination of homomorphic encryption, a verifiable secret redistribution scheme, and a mix network based on telescoping circuits. Our evaluation shows that Mycelium can answer a range of different questions from the medical literature with millions of devices.

Homomorphic encryption (HE) is considered as one of the most important primitives for privacy-preserving applications. However, an efficient approach to evaluate both polynomial and non-polynomial functions on encrypted data is still absent, which hinders the deployment of HE to real-life applications. To address this issue, we propose a practical framework PEGASUS. PEGASUS can efficiently switch back and forth between a packed CKKS ciphertext and FHEW ciphertexts without decryption, allowing us to evaluate arithmetic functions efficiently on the CKKS side, and to evaluate look-up tables on FHEW ciphertexts. Our FHEW → CKKS conversion algorithm is more practical than the existing methods. We improve the computational complexity from linear to sublinear. Moreover, the size of our conversion key is significantly smaller, e.g., reduced from 80 gigabytes to 12 megabytes. We present extensive benchmarks of PEGASUS, including sigmoid/ReLU/min/max/division, sorting and max-pooling. To further demonstrate the capability of PEGASUS, we developed two more applications. The first one is a private decision tree evaluation whose communication cost is about two orders of magnitude smaller than the previous HE-based approaches. The second one is a secure K-means clustering that is able to run on thousands of encrypted samples in minutes that outperforms the best existing system by 14 × – 20×. To the best of our knowledge, this is the first work that supports practical K-means clustering using HE in a single server setting.