Decentralized learning (DL) is an emerging machine learning paradigm where nodes collaboratively train models without a central server. However, the collaborative nature of DL makes it vulnerable to backdoor attacks, where a model is taught to behave normally on standard inputs while executing hidden, malicious actions when encountering data with specific triggers. Backdoor attacks in DL remain understudied and existing defenses often overlook DL constraints. We introduce Argus, a novel backdoor detection framework native to DL that requires neither a central coordinator nor prior knowledge of the trigger. In Argus, honest nodes locally analyze received model updates to identify potential backdoor triggers. Nodes then collectively share their triggers with their neighbors and use a structural similarity metric to separate true backdoors from false alarms induced by data heterogeneity. A key insight is that false positive triggers exhibit inconsistencies across participants while true positive ones show consistent patterns. Model updates that fail this collaborative test are rejected, and persistently malicious senders are eventually evicted. We provide the first theoretical convergence guarantees for a DL-specific backdoor detection mechanism, showing that filtering out suspicious model updates with high probability preserves a convergence rate comparable to standard DL. We implement and evaluate Argus on three standard datasets and against three state-of-the-art baselines. Across settings, Argus reduces attack success rates by up to 90 points compared to no defense, while preserving model utility within 5 percentage points of an omniscient oracle. Furthermore, the effectiveness of Argus compared to baselines improves as data heterogeneity increases.

Decentralized learning (DL) enables collaborative machine learning (ML) without a central server, making it suitable for settings where training data cannot be centrally hosted. We introduce Mosaic Learning, a DL framework that decomposes models into fragments and disseminates them independently across the network. Fragmentation reduces redundant communication across correlated parameters and enables more diverse information propagation without increasing communication cost. We theoretically show that Mosaic Learning (i) shows state-of-the-art worst-case convergence rate, and (ii) leverages parameter correlation in an ML model, improving contraction by reducing the highest eigenvalue of a simplified system. We empirically evaluate Mosaic Learning on four learning tasks and observe up to 12 percentage points higher node-level test accuracy compared to epidemic learning (EL), a state-of-the-art baseline. In summary, Mosaic Learning improves DL performance without sacrificing its utility or efficiency, and positions itself as a new DL standard.

As shown by Reliable Broadcast and Consensus, cooperation among a set of independent computing entities (sequential processes) is crucial in fault-tolerant distributed computing. Considering nprocess asynchronous message-passing systems where some processes may be Byzantine, this paper introduces a novel cooperation abstraction, Contention-Aware Cooperation (CAC). While Reliable Broadcast is a one-to- n cooperation abstraction and Consensus is an n-to- n cooperation abstraction, CAC is a d-to- n cooperation abstraction where d (1 <= d <= n) varies with each run and remains unknown to the processes. Correct processes accept the same set of l pairs v, i ( v is the value proposed by p(i)) from the d proposer processes, where 1 <= l <= d and (as d) l remains unknown to the processes (except in specific cases). Those l values are accepted one at a time, potentially in different orders at each process. In addition, CAC provides each process with an imperfect oracle that provides insights into the values that they may accept in the future. Interestingly, the CAC abstraction is particularly efficient in favorable circumstances, when the oracle becomes accurate, which processes can detect. To illustrate its practical utility, the paper details two applications leveraging CAC: a fast consensus implementation optimized for low contention (named Cascading Consensus), and a novel naming problem that can be solved under full asynchrony. All algorithms presented require signatures.

Decentralized Learning (DL) enables users to collaboratively train models without sharing raw data by iteratively averaging local updates with neighbors in a network graph. This setting is increasingly popular for its scalability and its ability to keep data local under user control. Strong privacy guarantees in DL are typically achieved through Differential Privacy (DP), with results showing that DL can even amplify privacy by disseminating noise across peer-to-peer communications. Yet in practice, the observed privacy-utility trade-off often appears worse than in centralized training, which may be due to limitations in current DP accounting methods for DL. In this paper, we show that recent advances in centralized DP accounting based on Matrix Factorization (MF) for analyzing temporal noise correlations can also be leveraged in DL. By generalizing existing MF results, we show how to cast both standard DL algorithms and common trust models into a unified formulation. This yields tighter privacy accounting for existing DP-DL algorithms and provides a principled way to develop new ones. To demonstrate the approach, we introduce MAFALDA-SGD, a gossip-based DL algorithm with user-level correlated noise that outperforms existing methods on synthetic and real-world graphs.

We address the problem of Reliable Broadcast in asynchronous message-passing systems with n nodes, of which up to t are malicious (faulty), in addition to a message adversary that can drop some of the messages sent by correct (non-faulty) nodes. We present a Message-Adversary-Tolerant Byzantine Reliable Broadcast (MBRB) algorithm that communicates O(|m| + n kappa) bits per node, where |m| represents the length of the application message and kappa = Omega(log n) is a security parameter . This communication complexity is optimal up to the parameter kappa. This significantly improves upon the state-of-the-art MBRB solution (Albouy, Frey, Raynal, and Taiani, TCS 2023), which incurs communication of O(n|m| + n(2)kappa) bits per node. Our solution sends at most 4n(2) messages overall, which is asymptotically optimal. Reduced communication is achieved by employing coding techniques that replace the need for all nodes to (re-)broadcast the entire application message m. Instead, nodes forward authenticated fragments of the encoding of m using an erasure-correcting code. Under the cryptographic assumptions of threshold signatures and vector commitments, and assuming n > 3t + 2d, where the adversary drops at most d messages per broadcast, our algorithm allows at least l = n - t - (1 + epsilon) d (for any arbitrarily low epsilon > 0) correct nodes to reconstruct m, despite missing fragments caused by the malicious nodes and the message adversary.

Virtual machines (VMs) raise several challenges for efficiently accommodating Cloud-gaming needs. The first challenge concerns VM boot time. Booting a VM can take several milliseconds to several seconds based on the operating system (OS) and the modules that must be loaded. Consequently, waiting for a VM to boot upon a user’s game-start request impacts the game startup latency and, thus, user experience. To mitigate this startup issue, Cloud-gaming providers maintain a pool of pre-initialized VMs, which leads to resource waste when the pool is not fully used. The second challenge concerns the flexibility of VMs regarding their memory size due to the lack of memory hotplug and unplug features. VMs cannot adapt their memory size to accommodate the heterogeneous memory requirements of existing video games. Concretely, pre-initialized VMs for a specific game cannot always be reused to host another game with different memory requirements. This means pre-initialized VMs must be configured with large memory sizes to host several games, leading to memory waste.In this paper, we bootstrap a discussion to steer away from a VM-centric Cloud-gaming infrastructure in favor of more resource-efficient infrastructures. To this end, we take a two-fold approach. First, we propose using containers as an efficient alternative to Windows VMs for video games. Our experiments show that the lightweight nature of containers achieves up to 10.6% energy-usage reduction without impacting the gaming experience. Second, we propose a novel GPU-sharing idea tailored for game containers. We discuss state-of-the-art GPU-sharing approaches, show their practical limitations for Cloud-gaming platforms, and present our envisioned GPU-sharing approach.

Decentralized learning (DL) is an emerging paradigm of collaborative machine learning that enables nodes in a network to train models collectively without sharing their raw data or relying on a central server. This paper introduces Zip-DL, a privacy-aware DL algorithm that leverages correlated noise to achieve robust privacy against local adversaries while ensuring efficient convergence at low communication costs. By progressively neutralizing the noise added during distributed averaging, Zip-DL combines strong privacy guarantees with high model accuracy. Its design requires only one communication round per gradient descent iteration, significantly reducing communication overhead compared to competitors. We establish theoretical bounds on both convergence speed and privacy guarantees. Moreover, extensive experiments demonstrating Zip-DL's practical applicability make it outperform state-of-the-art methods in the accuracy vs. vulnerability trade-off. Specifically, Zip-DL (i) reduces membership-inference attack success rates by up to 35% compared to baseline DL, (ii) decreases attack efficacy by up to 13% compared to competitors offering similar utility, and (iii) achieves up to 59% higher accuracy to completely nullify a basic attack scenario, compared to a state-of-the-art privacy-preserving approach under the same threat model. These results position Zip-DL as a practical and efficient solution for privacy-preserving decentralized learning in real-world applications.

Layer 2 rollups improve throughput and fees, but can reintroduce risk through operator discretion and information asymmetry. We ask which operator and governance designs produce ethically problematic user risk. We adapt Ethical Risk Analysis to rollup architectures, build a role-based taxonomy of decision authority and exposure, and pair the framework with two empirical signals, a cross sectional snapshot of 129 projects from L2BEAT and a hand curated incident set covering 2022 to 2025. We analyze mechanisms that affect risks to users funds, including upgrade timing and exit windows, proposer liveness and whitelisting, forced inclusion usability, and data availability choices. We find that ethical hazards rooted in L2 components control arrangements are widespread: instant upgrades without exit windows appear in about 86 percent of projects, and proposer controls that can freeze withdrawals in about 50 percent. Reported incidents concentrate in sequencer liveness and inclusion, consistent with these dependencies. We translate these findings into ethically grounded suggestions on mitigation strategies including technical components and governance mechanisms.

We address the problem of Reliable Broadcast in asynchronous message-passing systems with n nodes, of which up to t are malicious (faulty), in addition to a message adversary that can drop some of the messages sent by correct (non-faulty) nodes. We present a Message-Adversary-Tolerant Byzantine Reliable Broadcast (MBRB) algorithm that communicates an almost optimal amount of O(|m| + n(2)kappa) bits per node, where |m| represents the length of the application message and kappa = Omega(log n) is a security parameter. This improves upon the state-of-the-art MBRB solution (Albouy, Frey, Raynal, and Taiani, TCS 2023), which incurs communication of O(n|m| + n(2)kappa) bits per node. Our solution sends at most 4n(2) messages overall, which is asymptotically optimal. Reduced communication is achieved by employing coding techniques that replace the need for all nodes to (re-)broadcast the entire application message m. Instead, nodes forward authenticated fragments of the encoding of m using an erasure-correcting code. Under the cryptographic assumptions of PKI and collision-resistant hash, and assuming n > 3t+2d, where the adversary drops at most d messages per broadcast, our algorithm allows at least l = n - t - (1 + epsilon) d (for any epsilon > 0) correct nodes to reconstruct m, despite missing fragments caused by the malicious nodes and the message adversary.

End-to-end encrypted messaging applications such as Signal became widely popular thanks to their capability to ensure the confidentiality and integrity of online communication. While the highest security guarantees were long reserved for two-party communication, solutions for n-party communication remained either inefficient or less secure until the standardization of the MLS protocol (Messaging Layer Security). This new protocol offers an efficient way to provide end-to-end secure communication with the same guarantees originally offered by the Signal protocol for two-party communication. However, both solutions still rely on a centralized component for message delivery, called the Delivery Service in the MLS protocol. The centralization of the delivery service makes it an ideal target for attackers and threatens the availability of any protocol relying on MLS. In order to overcome this issue, we propose DiSCreet (Distributed delIvery Service with Context-awaRE coopEraTion), a design that allows clients to exchange protocol messages efficiently and without any intermediary. It uses a probabilistic reliable-broadcast mechanism to efficiently deliver messages and the Cascade Consensus protocol to handle messages requiring an agreement. Our solution strengthens the availability of the MLS protocol without compromising its security. We compare the theoretical performance of DiSCreet with another distributed solution, the DCGKA protocol, and detail the implementation of our solution.

In this paper, we examine how pre-processing and training methods impact on the performance of Lightweight CNNs through evaluations on MobileNetV3 with a spoofing detection head, dubbed "MobileNetV3-Spoof". Using the UniAttackData dataset from the 5th Face Anti-Spoofing Challenge@CVPR2024, which covers a broad spectrum of spoofing scenarios including deepfake and adversarial attack samples, we assess how well the model performs over different setups, including pre-trained models and models trained from scratch with or without initial face detection and alignment. Our results show that pre-processing steps significantly boost the model's ability to identify spoof samples, especially against complex attacks. Through detailed comparisons, we offer insights that could guide data curation and the creation of more effective and efficient antispoofing techniques suitable for real-world use in the era of digital face attacks. We make our code publicly available at: https://github. com/Inria- CENATAVTec/Assessing-Efficient- FAS-CVPR2024

This paper explores the territory that lies between best-effort Byzantine-Fault-Tolerant Conflict-free Replicated Data Types (BFT CRDTs) and totally ordered distributed ledgers, such as those implemented by Blockchains. It formally characterizes a novel class of distributed objects that only requires a First In First Out (FIFO) order on the object operations from each process (taken individually). The formalization leverages Mazurkiewicz traces to define legal sequences of operations and ensure both Strong Eventual Consistency (SEC) and Pipleline Consistency (PC). The paper presents a generic algorithm that implements this novel class of distributed objects both in a crash- and Byzantine setting. It also illustrates the practical interest of the proposed approach using four instances of this class of objects, namely money transfer, Petri nets, multi-sets, and concurrent work stealing dequeues.

With the growing breakthrough of deep learning-based face recognition, the development of lightweight models that achieve high accuracy with computational and memory efficiency has become paramount, especially for deployment on embedded domains. While Vision Transformers have shown significant promising results in various computer vision tasks, their adaptability to resource-constrained devices remains a significant challenge. This paper introduces SwiftFaceFormer, a new efficient, and lightweight family of face recognition models inspired by the hybrid SwiftFormer architecture. Our proposal not only retains the representational capacity of its predecessor but also introduces efficiency improvements, enabling enhanced face recognition performance at a fraction of the computational cost. We also propose to enhance the verification performance of our original most lightweight variant by using a training paradigm based on Knowledge Distillation. Through extensive experiments on several face benchmarks, the presented SwiftFaceFormer demonstrates high levels of accuracy compared to the original SwiftFormer model, and very competitive results with respect to state-of-the-art deep face recognition models, providing a suitable solution for real-time, on-device face recognition applications. Our code is available at https://github.com/Inria-CENATAV-Tec/SwiftFaceFormer.

This paper introduces a new asynchronous Byzantine-tolerant asset transfer system (cryptocurrency) with three noteworthy properties: quasi-anonymity, lightness, and consensus-freedom. Quasi-anonymity means no information is leaked regarding the receivers and amounts of the asset transfers. Lightness means that the underlying cryptographic schemes are succinct (i.e., they produce short-sized and quickly verifiable proofs) and each process only stores its own transfers while keeping communication cost as low as possible. Consensus-freedom means the system does not rely on a total order of asset transfers. The proposed algorithm is the first asset transfer system that simultaneously fulfills all these properties in the presence of asynchrony and Byzantine processes. To obtain them, the paper adopts a modular approach combining a new distributed object called “agreement proof” and well-known techniques such as commitments, universal accumulators, and zero-knowledge proofs.

This paper introduces ZIP-DL, a novel privacy-aware decentralized learning (DL) algorithm that exploits correlated noise to provide strong privacy protection against a local adversary while yielding efficient convergence guarantees for a low communication cost. The progressive neutralization of the added noise during the distributed aggregation process results in ZIP-DL fostering a high model accuracy under privacy guarantees. ZIP-DL further uses a single communication round between each gradient descent, thus minimizing communication overhead. We provide theoretical guarantees for both convergence speed and privacy guarantees, thereby making ZIP-DL applicable to practical scenarios. Our extensive experimental study shows that ZIP-DL significantly outperforms the state-of-the-art in terms of vulnerability/accuracy trade-off. In particular, ZIP-DL (i) reduces the efficacy of linkability attacks by up to 52 percentage points compared to baseline DL, (ii) improves accuracy by up to 37 percent w.r.t. the state-of-the-art privacy-preserving mechanism operating under the same threat model as ours, when configured to provide the same protection against membership inference attacks, and (iii) reduces communication by up to 10.5x against the same competitor for the same level of protection.

This paper considers the good-case latency of Byzantine Reliable Broadcast (BRB), i.e., the time taken by correct processes to deliver a message when the initial sender is correct. This time plays a crucial role in the performance of practical distributed systems. Although significant strides have been made in recent years on this question, progress has mainly focused on either asynchronous or randomized algorithms. By contrast, the good-case latency of deterministic synchronous BRB under a majority of Byzantine faults has been little studied. In particular, it was not known whether a goodcase latency below the worst-case bound of t + 1 rounds could be obtained. This work answers this open question positively and proposes a deterministic synchronous Byzantine reliable broadcast that achieves a good-case latency of max(2, t + 3 -- c) rounds, where t is the upper bound on the number of Byzantine processes and c the number of effectively correct processes.

We present SplitChain, a protocol intended to support the creation of scalable proof of stake and account-based blockchains without undermining decentralization and security. This is achieved by using sharding. SplitChain is designed to dynamically adapt the number of shards to the system load to avoid over-dimensioning issues encountered in static sharding-based solutions.

In this paper, we evaluate the performance and analyze the explainability of machine learning models boosted by feature selection in predicting COVID-19-positive cases from self-reported information. In essence, this work describes a methodology to identify COVID-19 infections that considers the large amount of information collected by the University of Maryland Global COVID-19 Trends and Impact Survey (UMD-CTIS). More precisely, this methodology performs a feature selection stage based on the recursive feature elimination (RFE) method to reduce the number of input variables without compromising detection accuracy. A tree-based supervised machine learning model is then optimized with the selected features to detect COVID-19-active cases. In contrast to previous approaches that use a limited set of selected symptoms, the proposed approach builds the detection engine considering a broad range of features including self-reported symptoms, local community information, vaccination acceptance, and isolation measures, among others. To implement the methodology, three different supervised classifiers were used: random forests (RF), light gradient boosting (LGB), and extreme gradient boosting (XGB). Based on data collected from the UMD-CTIS, we evaluated the detection performance of the methodology for four countries (Brazil, Canada, Japan, and South Africa) and two periods (2020 and 2021). The proposed approach was assessed in terms of various quality metrics: F1-score, sensitivity, specificity, precision, receiver operating characteristic (ROC), and area under the ROC curve (AUC). This work also shows the normalized daily incidence curves obtained by the proposed approach for the four countries. Finally, we perform an explainability analysis using Shapley values and feature importance to determine the relevance of each feature and the corresponding contribution for each country and each country/year.