Showing papers by "George Danezis published in 2019"

LOGAN: Membership Inference Attacks Against Generative Models

Abstract: Generative models estimate the underlying distribution of a dataset to generate realistic samples according to that distribution. In this paper, we present the first membership inference attacks against generative models: given a data point, the adversary determines whether or not it was used to train the model. Our attacks leverage Generative Adversarial Networks (GANs), which combine a discriminative and a generative model, to detect overfitting and recognize inputs that were part of training datasets, using the discriminator's capacity to learn statistical differences in distributions. We present attacks based on both white-box and black-box access to the target model, against several state-of-the-art generative models, over datasets of complex representations of faces (LFW), objects (CIFAR-10), and medical images (Diabetic Retinopathy). We also discuss the sensitivity of the attacks to different training parameters, and their robustness against mitigation strategies, finding that defenses are either ineffective or lead to significantly worse performances of the generative models in terms of training stability and/or sample quality.

SoK: Consensus in the Age of Blockchains

Abstract: The core technical component of blockchains is consensus: how to reach agreement among a distributed network of nodes. A plethora of blockchain consensus protocols have been proposed---ranging from new designs, to novel modifications and extensions of consensus protocols from the classical distributed systems literature. The inherent complexity of consensus protocols and their rapid and dramatic evolution makes it hard to contextualize the design landscape. We address this challenge by conducting a systematization of knowledge of blockchain consensus protocols. After first discussing key themes in classical consensus protocols, we describe: (i) protocols based on proof-of-work; (ii) proof-of-X protocols that replace proof-of-work with more energy-efficient alternatives; and (iii) hybrid protocols that are compositions or variations of classical consensus protocols. This survey is guided by a systematization framework we develop, to highlight the various building blocks of blockchain consensus design, along with a discussion on their security and performance properties. We identify research gaps and insights for the community to consider in future research endeavours.

‘Internet of Things’: How Abuse is Getting Smarter

Abstract: From home thermostats you can control from your car, to home assistants ready to organise your diary at a spoken word, technology is playing a more central role in our daily lives. However, while networked home devices provide many advantages, they also offer abusers an abundance of opportunities to control, harass and stalk their victims. The Gender and Internet of Things project at University College London has been investigating how these devices are being misused, and what support survivors and services need to navigate these emerging risks.

Replay Attacks and Defenses Against Cross-shard Consensus in Sharded Distributed Ledgers

Abstract: We present a family of replay attacks against sharded distributed ledgers, that target cross-shard consensus protocols, such as the recently proposed Chainspace and Omniledger. They allow an attacker, with network access only, to double-spend or lock resources with minimal efforts. The attacker can act independently without colluding with any nodes, and succeed even if all nodes are honest; most of the attacks can also exhibit themselves as faults under periods of asynchrony. These attacks are effective against both shard-led and client-led cross-shard consensus approaches. Finally, we present Byzcuit - a new cross-shard consensus protocol that is immune to those attacks. We implement a prototype of Byzcuit and evaluate it on a real cloud-based testbed, showing that our defenses impact performance minimally, and overall performance surpasses previous works.

‘Internet of Things’: How abuse is getting smarter

Abstract: From home thermostats you can control from your car, to home assistants ready to organise your diary at a spoken word, technology is playing a more central role in our daily lives. However, while networked home devices provide many advantages, they also offer abusers an abundance of opportunities to control, harass and stalk their victims. The Gender and Internet of Things project at University College London has been investigating how these devices are being misused, and what support survivors and services need to navigate these emerging risks.

No Right to Remain Silent: Isolating Malicious Mixes

Abstract: Mix networks are a key technology to achieve network anonymity and private messaging, voting and database lookups. However, simple mix network designs are vulnerable to malicious mixes, which may drop or delay packets to facilitate traffic analysis attacks. Mix networks with provable robustness address this drawback through complex and expensive proofs of correct shuffling but come at a great cost and make limiting or unrealistic systems assumptions. We present Miranda, an efficient mix-net design, which mitigates active attacks by malicious mixes. Miranda uses both the detection of corrupt mixes, as well as detection of faults related to a pair of mixes, without detection of the faulty one among the two. Each active attack -- including dropping packets -- leads to reduced connectivity for corrupt mixes and reduces their ability to attack, and, eventually, to detection of corrupt mixes. We show, through experiments, the effectiveness of Miranda, by demonstrating how malicious mixes are detected and that attacks are neutralized early.

SybilQuorum: Open Distributed Ledgers Through Trust Networks

Abstract: The Sybil attack plagues all peer-to-peer systems, and modern open distributed ledgers employ a number of tactics to prevent it from proof of work, or other resources such as space, stake or memory, to traditional admission control in permissioned settings. With SybilQuorum we propose an alternative approach to securing an open distributed ledger against Sybil attacks, and ensuring consensus amongst honest participants, leveraging social network based Sybil defences. We show how nodes expressing their trust relationships through the ledger can bootstrap and operate a value system, and general transaction system, and how Sybil attacks are thwarted. We empirically evaluate our system as a secure Federated Byzantine Agreement System, and extend the theory of those systems to do so.