Among the many technical challenges to enforcing AI regulations, one crucial yet underexplored problem is the risk of audit manipulation. This manipulation occurs when a platform deliberately alters its answers to a regulator to pass an audit without modifying its answers to other users. In this paper, we introduce a novel approach to manipulation-proof auditing by taking into account the auditor's prior knowledge of the task solved by the platform. We first demonstrate that regulators must not rely on public priors (e.g. a public dataset), as platforms could easily fool the auditor in such cases. We then formally establish the conditions under which an auditor can prevent audit manipulations using prior knowledge about the ground truth. Finally, our experiments with two standard datasets illustrate the maximum level of unfairness a platform can hide before being detected as malicious. Our formalization and generalization of manipulation-proof auditing with a prior opens up new research directions for more robust fairness audits.

Decentralized learning (DL) enables collaborative learning without a server and without training data leaving the users' devices. However, the models shared in DL can still be used to infer training data. Conventional defenses such as differential privacy and secure aggregation fall short in effectively safeguarding user privacy in DL, either sacrificing model utility or efficiency. We introduce Shatter, a novel DL approach in which nodes create virtual nodes (VNs) to disseminate chunks of their full model on their behalf. This enhances privacy by (i) preventing attackers from collecting full models from other nodes, and (ii) hiding the identity of the original node that produced a given model chunk. We theoretically prove the convergence of Shatter and provide a formal analysis demonstrating how Shatter reduces the efficacy of attacks compared to when exchanging full models between nodes. We evaluate the convergence and attack resilience of Shatter with existing DL algorithms, with heterogeneous datasets, and against three standard privacy attacks. Our evaluation shows that Shatter not only renders these privacy attacks infeasible when each node operates 16 VNs but also exhibits a positive impact on model utility compared to standard DL. In summary, Shatter enhances the privacy of DL while maintaining the utility and efficiency of the model.

Existing work in fairness auditing assumes that each audit is performed independently. In this paper, we consider multiple agents working together, each auditing the same platform for different tasks. Agents have two levers: their collaboration strategy, with or without coordination beforehand, and their strategy for sampling appropriate data points. We theoretically compare the interplay of these levers. Our main findings are that (i) collaboration is generally beneficial for accurate audits, (ii) basic sampling methods often prove to be effective, and (iii) counter-intuitively, extensive coordination on queries often deteriorates audits accuracy as the number of agents increases. Experiments on three large datasets confirm our theoretical results. Our findings motivate collaboration during fairness audits of platforms that use ML models for decision-making.

Federated Unlearning (FU) aims to delete specific training data from an ML model trained using Federated Learning (FL). However, existing FU methods suffer from inefficiencies due to the high costs associated with gradient recomputation and storage. This paper presents QuickDrop, an original and efficient FU approach designed to overcome these limitations. During model training, each client uses QuickDrop to generate a compact synthetic dataset, serving as a compressed representation of the gradient information utilized during training. This synthetic dataset facilitates fast gradient approximation, allowing rapid downstream unlearning at minimal storage cost. To unlearn some knowledge from the trained model, QuickDrop clients execute stochastic gradient ascent with samples from the synthetic datasets instead of the training dataset. The tiny volume of synthetic data significantly reduces computational overhead compared to conventional FU methods. Evaluations with three standard datasets and five baselines show that, with comparable accuracy guarantees, QuickDrop reduces the unlearning duration by 463× compared to retraining the model from scratch and 65 -- 218× compared to FU baselines. QuickDrop supports both class- and client-level unlearning, multiple unlearning requests, and relearning of previously erased data.

The ability of a peer-to-peer (P2P) system to effectively host decentralized applications often relies on the availability of a peer-sampling service, which provides each participant with a random sample of other peers. Despite the practical effectiveness of existing peer samplers, their ability to produce random samples within a reasonable time frame remains poorly understood from a theoretical standpoint. This paper contributes to bridging this gap by introducing PeersWap,a peer-sampling protocol with provable randomness guarantees. We establish execution time bounds for PeerSwap, demonstrating its ability to scale effectively with the network size. We prove that PeerSwap maintains the fixed structure of the communication graph while allowing sequential peer position swaps within this graph. We do so by showing that PeerSwap is a specific instance of an interchange process, a renowned model for particle movement analysis. Leveraging this mapping, we derive execution time bounds, expressed as a function of the network size n. Depending on the network structure, this time can be as low as a polylogarithmic function of n, highlighting the efficiency of PeerSwap. We implement PeerSwap and conduct numerical evaluations using regular graphs with varying connectivity and containing up to 32768(215) peers. Our evaluation demonstrates that PeerSwap quickly provides peers with uniform random samples of other peers.

With the proliferation of Internet of Things (IoT) ecosystems, traditional resource orchestration mechanisms, executed on fog devices, encounter significant scalability, reliability and security challenges. To tackle these challenges, recent decentralized algorithms in Fog-IoT use Distributed Ledger Technologies to orchestrate resources and payments between peers. However, while distributed ledgers provide many desirable properties, their consensus mechanism introduces a performance bottleneck. This paper introduces Light-HIDRA, a consensus-less and decentralized resource orchestration system for Fog-IoT environments. At its core, Light-HIDRA uses Byzantine Reliable Broadcast (BRB) to coordinate actions without centralized control, therefore drastically reducing communication overhead and latency compared to consensus-based solutions. Light-HIDRA coordinates the scheduling and execution of workloads, and securely manages the payments that peers receive for dedicating resources to workloads. Light-HIDRA further increases performance and reduces overhead by grouping peers into distinct domains. We conduct an in-depth analysis of the protocol’s security properties, investigating its efficiency and robustness in diverse situations. We evaluate the performance of Light-HIDRA, highlighting its performance over HIDRA, a state-of-the-art baseline that uses smart contracts. Our experiments demonstrate that Light-HIDRA reduces the bandwidth usage by up to 57x, the latency of workload offloading by up to 142x, and shows superior throughput compared to HIDRA.

The popularity of blockchain technology has bootstrapped many “Web3” applications, e.g., Ethereum and IPFS, that apply distributed ledger technology to store transactions. The amount of transactions generated and stored in such Web3 applications is significant and, in its raw form, usually not searchable by users. Existing Web3 transaction indexing and search engines are predominantly centralized and, therefore, can manipulate search results or censor particular queries. With the proliferation of Web3 transactions and applications, a decentralized and censorship-resistant search primitive is becoming essential. We present DeScan, a decentralized and censorship-resistant indexing and search engine for Web3. Users index their local Web3 transactions using custom rules that output triplets. Generated triplets are bundled in a distributed transaction graph that is searchable by other users. To coordinate search and distribute the storage of the transaction graph over peers in the network, we build upon a Skip Graph (SG) data structure. Since the Skip Graph does not provide any resilience against adversarial peers that censor searches, we propose four modifications to improve its robustness. We implement DeScan and conduct experiments with up to 12 800 peers and 10 million Ethereum transactions. Our experiments show that DeScan with our modifications enabled can tolerate 20% adversarial peers and 35% unresponsive peers without disruption. Moreover, we find that searches in DeScan are usually completed well within a second, even when the network grows. Finally, we show that storage and network costs are evenly distributed amongst peers as the network grows.

We present Epidemic Learning (EL), a simple yet powerful decentralized learning (DL) algorithm that leverages changing communication topologies to achieve faster model convergence compared to conventional DL approaches. At each round of EL, each node sends its model updates to a random sample of $s$ other nodes (in a system of $n$ nodes). We provide an extensive theoretical analysis of EL, demonstrating that its changing topology culminates in superior convergence properties compared to the state-of-the-art (static and dynamic) topologies. Considering smooth non-convex loss functions, the number of transient iterations for EL, i.e., the rounds required to achieve asymptotic linear speedup, is in $O(n^3/s^2)$ which outperforms the best-known bound $O(n^3)$ by a factor of $s^2$, indicating the benefit of randomized communication for DL. We empirically evaluate EL in a 96-node network and compare its performance with state-of-the-art DL approaches. Our results illustrate that EL converges up to $ 1.7\times$ quicker than baseline DL algorithms and attains $2.2 $\% higher accuracy for the same communication volume.

“Big Tech” companies provide digital services used by billions of people. Recent developments, however, have shown that these companies often abuse their unprecedented market dominance for selfish interests. Meanwhile, decentralized applications without central authority are gaining traction. Decentralized applications critically depend on its users working together. Ensuring that users do not consume too many resources without reciprocating is a crucial requirement for the sustainability of such applications. We present ConTrib, a universal mechanism to maintain fairness in decentralized applications by accounting the work performed by peers. In ConTrib, participants maintain a personal ledger with tamper-evident records. A record describes some work performed by a peer and links to other records. Fraud in ConTrib occurs when a peer illegitimately modifies one of the records in its personal ledger. This is detected through the continuous exchange of random records between peers and by verifying the consistency of incoming records against known ones. Our simple fraud detection algorithm is highly scalable, tolerates significant packet loss, and exhibits relatively low fraud detection times. We experimentally show that fraud is detected within seconds and with low bandwidth requirements. To demonstrate the applicability of our work, we deploy ConTrib in the Tribler file-sharing application and successfully address free-riding behaviour. This two-year trial has resulted in over 160 million records, created by more than 94’000 users.

Abstract Permissioned blockchains are increasingly being used as a solution to record transactions between companies. Several use cases that leverage permissioned blockchains focus on the representation and management of real-world assets. Since the number of incompatible blockchains is quickly growing, there is an increasing need for a universal mechanism to exchange, or trade, digital assets between these isolated platforms. There currently is no universal mechanism for inter-blockchain asset exchange without a requirement for trusted authorities that coordinate the trade. We address this shortcoming and present XChange, a universal mechanism for asset exchange between permissioned blockchains. To achieve universality and to avoid trusted authorities that coordinate a trade, XChange does not provide atomic guarantees but leverages risk mitigation strategies to reduce value at stake. Our mechanism records the specifications and progression of each trade within records on a distributed log. XChange reduces the economic gains of adversaries by bounding the total amount of fraud they can commit at any time. After having committed fraud, an adversary is forced to finish its ongoing trades before it can engage in new trades. We first present a four-phased protocol that coordinates an asset exchange between two traders. We then outline how trade records can be stored on TrustChain, which is a lightweight distributed ledger specifically built for the tamper-proof storage of data elements. We implement XChange and conduct experiments. Our experiments demonstrate that XChange is capable of reducing the economic gains of adversaries by more than 99.9% when replaying a real-world trading dataset. A deployment on low-resource devices reveals that the latency added to a trade by XChange is only 493 milliseconds. Finally, our scalability evaluation shows that XChange achieves over 1’000 trades per second and that its throughput, in terms of trades per second, scales linearly with the system load.

Matchmaking is a core enabling element in peer-to-peer markets. To date, matchmaking is predominantly performed by proprietary algorithms, fully controlled by market operators. This raises fairness concerns as market operators effectively can hide, prioritize, or delay the orders of specific users. Blockchain technology has been proposed as an alternative for fair matchmaking without a trusted operator but is still vulnerable to specific fairness attacks.