arXiv subjects

Jérémie Decouchant

Publications and source records attributed to Jérémie Decouchant.

At least 19 recordsLinked to original sources

arXiv2026

Absorbed in Inertia: Activation Analysis for Computer-Use Agents

Computer-use agents have become increasingly capable of executing tasks on live desktops through natural-language instructions, based on trajectories of screenshots, actions, and reasoning. We discover that they can stealthily exhibit inertia, in which they repeat fruitless actions despite recognizing that these actions are ineffective. We hypothesize that inertia is reflected in the agent's internal state, i.e., the activation values of the agent's underlying model, and propose a protocol to measure the relationship between the two. Extensive analysis of high-dimensional activation states shows that inertia corresponds to an absorbing region of activation space, where activation values become stale across actions and even after attempts to steer them. We conjecture that drastically changing the agents' activations by re-initializing them is necessary to escape inertia. Specifically, we propose R$^3$ (Reset, Reroute, Restore), which temporarily resets the agent's context trajectory to escape the absorbing region and then restores the historical context to effectively complete the task. Our approach yields 17-55% lower measured inertia across models relative to unmodified agents. These results suggest that changing the context can interrupt recurrence more effectively than directly steering the resulting activations. Our code is available at https://anonymous.4open.science/r/vlm-agent-defense-D076

cs.AI↗

arXiv2026

Mitigating Front-Running Attacks through Fair and Resilient Transaction Dissemination

In modern blockchains, efficient, fair, and faulttolerant information dissemination is critical for performance and security. Several stages of the transaction lifecycle are affected, from the creation and dissemination of transactions to the dissemination of blocks in the consensus layer. Mempool protocols, such as L{Ø}, already address some of modern blockchains' security threats. However, others remain unless fairness is embedded most fundamentally in the dissemination layer used by these protocols to share transactions and blocks and to reconcile mempools. This paper introduces HERMES, a novel dissemination protocol for mempools that leverages robust minimal structures to optimize data propagation, balance load, and ensure fairness despite faults and Byzantine actors. Specifically, we address front-running attacks by randomizing the choice of an overlay structure while forcing nodes to prove adherence to the mempools' dissemination policies and the random choice made. Experimental results show significant performance improvements compared to traditional broadcast protocols for both permissioned and permissionless blockchains.

cs.DC↗

arXiv2026

The Blockchain Execution Dilemma: Optimizing Revenue XOR Fair Ordering

Successive consensus algorithm generations displaced the blockchain performance bottleneck to the execution layer. Recent works address this by parallelizing non-conflicting transactions. Historically, execution followed consensus-level ordering, a practice susceptible to Maximal Extractable Value (MEV) exploitation. Conversely, recent academic proposals introduce rigid fair-ordering protocols that severely restrict transaction reordering. Parallel execution frameworks optimize the sequencing layer, located between consensus and execution, assembling transaction batches from committed blocks and transmitting them to workers to maximize execution parallelism and realized fees. To achieve this, current sequencing implementations may defer transactions, but they do not maintain order fairness. Preserving these properties currently requires sequential execution, drastically reducing profitability and performance. In this work, we address the tension between validator profit and order fairness using a dynamic optimization framework. We introduce a blockchain-agnostic model for transaction sequencing in a continuous setting where block sequencing and execution run concurrently. When sequencing cannot complete within the available time window, our framework dynamically returns its best intermediate result. Within this framework, we propose an anytime genetic algorithm that utilizes gas prices, object sets, and predicted execution times to optimize schedules, with an option to maintain fair ordering. Evaluating our approach on real-world Sui and Ethereum datasets shows it increases validator profit by ~15% and accelerates congestion relief speed by up to 58%. Furthermore, fair-ordering constraints can reduce validator profit by 50% to 60% during high congestion, providing the first evidence that enforcing strict fair ordering might effectively nullify the advantages of advanced sequencing.

cs.DC↗

arXiv2026

Diffusion LLMs as Targets and Adversaries: Mechanistic Safety Exploits

Diffusion Large Language Models (DLLMs) replace autoregressive next-token prediction with iterative parallel denoising, yet their internal safety mechanisms remain poorly understood. In this work, we investigate DLLMs both as targets and as adversaries, exposing mechanistic vulnerabilities in diffusion-based alignment. We first show that safety alignment in DLLMs remains sparse and transferable across architectures. DLLMs initialized from autoregressive predecessors inherit the same mechanistic safety footprint as their source models, enabling transfer attacks via direct safety neuron mapping and pruning. Self-pruning increases attack success rates (ASR) from 2.6% to 73.8% on LLaDA and from 1.9% to 86.6% on Dream, while transfer pruning from Qwen2.5 increases ASR from 1.9% to 73.2% on Dream and from 7.0% to 86.3% on Fast-dLLM. Building on these findings, we introduce SN-Guided Diffusion, a fully offline black-box jailbreak framework that steers the diffusion process away from safety-triggering regions using a weighted safety neuron loss, which achieves near-perfect prompt separability (AUROC = 1.0 for benign-vs-jailbreak discrimination). Across multiple open and proprietary targets, our method achieves a transfer ASR of up to 77.1% on Llama-3-8B-Instruct, 86.9% on Qwen2.5-7B-Instruct, and 74.3% against Gemini-2.5-Flash-Lite, while requiring only 20 generation episodes per prompt. Compared to prior jailbreaking frameworks, our method achieves competitive transferability with orders-of-magnitude lower generation cost. Our codebase is available at https://github.com/ellyoana/sn-guided-diffusion.

cs.LG↗

arXiv2026

Tilikum: Transaction Fair Ordering on a DAG without Weak Edges

Decentralized Finance (DeFi) applications rely heavily on the order in which transactions are executed, making them susceptible to reordering attacks that enable adversaries to extract Blockchain Extractable Value (BEV). While linear blockchain systems such as Ethereum have inspired extensive research into fair ordering mechanisms, DAG-based consensus protocols have remained largely unprotected despite their growing adoption for scalability and performance. In this paper, we introduce Tilikum, a DAG-based ledger protocol that ensures fair transaction ordering without relying on weak edges. Tilikum achieves ordering linearizability by leveraging median-based timestamp aggregation, or batch order fairness, while maintaining low data redundancy and robust garbage collection. We implemented Tilikum in Rust and evaluated it against representative baselines, namely Narwhal/Tusk, Pompē, Themis and FairDAG. Our results show that Tilikum achieves up to $39\times$ higher throughput than other fair-ordering baselines, while fully blocking state-of-the-art DAG-specific reordering attacks.

cs.CR↗

arXiv2026

Robust and Automated Reconfiguration of Byzantine Wide-Area Replication

Distributed systems handle adversarial nodes through redundancy, which imposes a significant performance overhead. In blockchain systems, Byzantine fault-tolerant state-machine replication (BFT-SMR) is the replicated service that totally orders client transactions before execution. While prior research has primarily focused on designing novel consensus algorithms with improved performance, recent studies have shown that further gains can be achieved through configuration optimization. More precisely, replicas can monitor network latency to dynamically assign the leader role and tune voting weights, thereby improving consensus performance. However, we identify three vulnerabilities in this process that Byzantine nodes can exploit. To address these weaknesses, we propose Beware, a reconfiguration framework that filters out falsified latency reports, computes robust weight distributions, and applies machine learning to converge towards Byzantine-resilient configurations. Our evaluation shows that Beware reduces consensus latency by up to 45% compared to existing solutions.

cs.DC↗

arXiv2026

Herring: Parallel Batch-Order-Fairness on DAG-based Blockchain Consensus

Transaction ordering attacks extract billions of dollars annually from decentralized finance users in the form of Maximal Extractable Value (MEV). Byzantine Fault-Tolerant (BFT) consensus protocols guarantee total order but place no constraint on how that order is chosen, leaving the door open for adversarial reordering. Batch-order-fairness (batch-OF) protocols close this gap, but existing designs pay a steep performance price for this guarantee. Leader-based protocols such as Themis concentrate all fairness decisions at a single replica, while recent DAG-based proposals FairDAG and DAG of DAGs (DoD) force their fairness layer into strictly serial execution despite running on multi-proposer DAGs. We present Herring, the first $γ$-batch-OF DAG BFT protocol whose fairness layer parallelizes the dominant graph construction cost across committed subdags. Herring combines post-consensus graph construction with explicit missing edge resolution piggybacked on the DAG's reliable broadcast layer, a pairing that turns fair ordering from a per-round serial bottleneck into a CPU-bound task. We also uncover previously unreported liveness vulnerabilities in both FairDAG-RL and DoD that a malicious client can trigger to halt the fairness layer indefinitely, and propose patches that we integrate into our reimplementations. We implement Herring on top of the Rust implementation of Narwhal \& Tusk and evaluate it against FairDAG-RL, DoD-W, and Themis. Herring tracks the throughput of Narwhal \& Tusk closely up to roughly $10{,}000$\,tx/s, achieves roughly $90\%$ higher saturation throughput than FairDAG-RL and $100\%$ higher than DoD-W, and substantially reduces execution latency at saturation.

cs.DC↗

arXiv2026

Aergia: Leveraging Heterogeneity in Federated Learning Systems

Federated Learning (FL) is a popular approach for distributed deep learning that prevents the pooling of large amounts of data in a central server. FL relies on clients to update a global model using their local datasets. Classical FL algorithms use a central federator that, for each training round, waits for all clients to send their model updates before aggregating them. In practical deployments, clients might have different computing powers and network capabilities, which might lead slow clients to become performance bottlenecks. Previous works have suggested to use a deadline for each learning round so that the federator ignores the late updates of slow clients, or so that clients send partially trained models before the deadline. To speed up the training process, we instead propose Aergia, a novel approach where slow clients (i) freeze the part of their model that is the most computationally intensive to train; (ii) train the unfrozen part of their model; and (iii) offload the training of the frozen part of their model to a faster client that trains it using its own dataset. The offloading decisions are orchestrated by the federator based on the training speed that clients report and on the similarities between their datasets, which are privately evaluated thanks to a trusted execution environment. We show through extensive experiments that Aergia maintains high accuracy and significantly reduces the training time under heterogeneous settings by up to 27% and 53% compared to FedAvg and TiFL, respectively.

cs.LG↗

arXiv2026

Dynamic Topology Optimization for Non-IID Data in Decentralized Learning

Decentralized learning (DL) enables a set of nodes to train a model collaboratively without central coordination, offering benefits for privacy and scalability. However, DL struggles to train a high accuracy model when the data distribution is non-independent and identically distributed (non-IID) and when the communication topology is static. To address these issues, we propose Morph, a topology optimization algorithm for DL. In Morph, nodes adaptively choose peers for model exchange based on maximum model dissimilarity. Morph maintains a fixed in-degree while dynamically reshaping the communication graph through gossip-based peer discovery and diversity-driven neighbor selection, thereby improving robustness to data heterogeneity. Experiments on CIFAR-10 and FEMNIST with up to 100 nodes show that Morph consistently outperforms static and epidemic baselines, while closely tracking the fully connected upper bound. On CIFAR-10, Morph achieves a relative improvement of 1.12x in test accuracy compared to the state-of-the-art baselines. On FEMNIST, Morph achieves an accuracy that is 1.08x higher than Epidemic Learning. Similar trends hold for 50 node deployments, where Morph narrows the gap to the fully connected upper bound within 0.5 percentage points on CIFAR-10. These results demonstrate that Morph achieves higher final accuracy, faster convergence, and more stable learning as quantified by lower inter-node variance, while requiring fewer communication rounds than baselines and no global knowledge.

cs.LG↗

arXiv2025

Balancing Fairness and Performance in Multi-User Spark Workloads with Dynamic Scheduling (extended version)

Apache Spark is a widely adopted framework for large-scale data processing. However, in industrial analytics environments, Spark's built-in schedulers, such as FIFO and fair scheduling, struggle to maintain both user-level fairness and low mean response time, particularly in long-running shared applications. Existing solutions typically focus on job-level fairness which unintentionally favors users who submit more jobs. Although Spark offers a built-in fair scheduler, it lacks adaptability to dynamic user workloads and may degrade overall job performance. We present the User Weighted Fair Queuing (UWFQ) scheduler, designed to minimize job response times while ensuring equitable resource distribution across users and their respective jobs. UWFQ simulates a virtual fair queuing system and schedules jobs based on their estimated finish times under a bounded fairness model. To further address task skew and reduce priority inversions, which are common in Spark workloads, we introduce runtime partitioning, a method that dynamically refines task granularity based on expected runtime. We implement UWFQ within the Spark framework and evaluate its performance using multi-user synthetic workloads and Google cluster traces. We show that UWFQ reduces the average response time of small jobs by up to 74% compared to existing built-in Spark schedulers and to state-of-the-art fair scheduling algorithms.

cs.DC↗

arXiv2025

NEMO: Faster Parallel Execution for Highly Contended Blockchain Workloads (Full version)

Following the design of more efficient blockchain consensus algorithms, the execution layer has emerged as the new performance bottleneck of blockchains, especially under high contention. Current parallel execution frameworks either rely on optimistic concurrency control (OCC) or on pessimistic concurrency control (PCC), both of which see their performance decrease when workloads are highly contended, albeit for different reasons. In this work, we present NEMO, a new blockchain execution engine that combines OCC with the object data model to address this challenge. NEMO introduces four core innovations: (i) a greedy commit rule for transactions using only owned objects; (ii) refined handling of dependencies to reduce re-executions; (iii) the use of incomplete but statically derivable read/write hints to guide execution; and (iv) a priority-based scheduler that favors transactions that unblock others. Through simulated execution experiments, we demonstrate that NEMO significantly reduces redundant computation and achieves higher throughput than representative approaches. For example, with 16 workers NEMO's throughput is up to 42% higher than the one of Block-STM, the state-of-the-art OCC approach, and 61% higher than the pessimistic concurrency control baseline used.

cs.DC↗

arXiv2025

StarveSpam: Mitigating Spam with Local Reputation in Permissionless Blockchains

Spam poses a growing threat to blockchain networks. Adversaries can easily create multiple accounts to flood transaction pools, inflating fees and degrading service quality. Existing defenses against spam, such as fee markets and staking requirements, primarily rely on economic deterrence, which fails to distinguish between malicious and legitimate users and often exclude low-value but honest activity. To address these shortcomings, we present StarveSpam, a decentralized reputation-based protocol that mitigates spam by operating at the transaction relay layer. StarveSpam combines local behavior tracking, peer scoring, and adaptive rate-limiting to suppress abusive actors, without requiring global consensus, protocol changes, or trusted infrastructure. We evaluate StarveSpam using real Ethereum data from a major NFT spam event and show that it outperforms existing fee-based and rule-based defenses, allowing each node to block over 95% of spam while dropping just 3% of honest traffic, and reducing the fraction of the network exposed to spam by 85% compared to existing rule-based methods. StarveSpam offers a scalable and deployable alternative to traditional spam defenses, paving the way toward more resilient and equitable blockchain infrastructure.

cs.CR↗

arXiv2025

Go With The Flow: Churn-Tolerant Decentralized Training of Large Language Models

Motivated by the emergence of large language models (LLMs) and the importance of democratizing their training, we propose GWTF, the first crash tolerant practical decentralized training framework for LLMs. Differently from existing distributed and federated training frameworks, GWTF enables the efficient collaborative training of a LLM on heterogeneous clients that volunteer their resources. In addition, GWTF addresses node churn, i.e., clients joining or leaving the system at any time, and network instabilities, i.e., network links becoming unstable or unreliable. The core of GWTF is a novel decentralized flow algorithm that finds the most effective routing that maximizes the number of microbatches trained with the lowest possible delay. We extensively evaluate GWTF on GPT-like and LLaMa-like models and compare it against the prior art. Our results indicate that GWTF reduces the training time by up to 45% in realistic and challenging scenarios that involve heterogeneous client nodes distributed over 10 different geographic locations with a high node churn rate.

cs.LG↗

arXiv2025

Share Secrets for Privacy: Confidential Forecasting with Vertical Federated Learning

Vertical federated learning (VFL) is a promising area for time series forecasting in many applications, such as healthcare and manufacturing. Critical challenges to address include data privacy and over-fitting on small and noisy datasets during both training and inference. Additionally, such forecasting models must scale well with the number of parties while ensuring strong convergence and low-tuning complexity. We address these challenges and propose ``Secret-shared Time Series Forecasting with VFL'' (STV), a novel framework with the following key features: i) a privacy-preserving algorithm for forecasting with SARIMAX and autoregressive trees on vertically-partitioned data; ii) decentralised forecasting using secret sharing and multi-party computation; and iii) novel N-party algorithms for matrix multiplication and inverse operations for exact parameter optimization, giving strong convergence with minimal tuning complexity. We evaluate on six representative datasets from public and industry-specific contexts. Results demonstrate that STV's forecasting accuracy is comparable to those of centralized approaches. Our exact optimization outperforms centralized methods, including state-of-the-art diffusion models and long-short-term memory, by 23.81% on forecasting accuracy. We also evaluate scalability by examining the communication costs of exact and iterative optimization to navigate the choice between the two. STV's code and supplementary material is available online: https://github.com/adis98/STV.

cs.LG↗

arXiv2025

Tolerating Disasters with Hierarchical Consensus

Geo-replication provides disaster recovery after catastrophic accidental failures or attacks, such as fires, blackouts or denial-of-service attacks to a data center or region. Naturally distributed data structures, such as Blockchains, when well designed, are immune against such disruptions, but they also benefit from leveraging locality. In this work, we consolidate the performance of geo-replicated consensus by leveraging novel insights about hierarchical consensus and a construction methodology that allows creating novel protocols from existing building blocks. In particular we show that cluster confirmation, paired with subgroup rotation, allows protocols to safely operate through situations where all members of the global consensus group are Byzantine. We demonstrate our compositional construction by combining the recent HotStuff and Damysus protocols into a hierarchical geo-replicated blockchain with global durability guarantees. We present a compositionality proof and demonstrate the correctness of our protocol, including its ability to tolerate cluster crashes. Our protocol -ORION 1 -achieves a 20% higher throughput than GeoBFT, the latest hierarchical Byzantine Fault-Tolerant (BFT) protocol.

cs.DC↗

arXiv2025

Taming Double-Spending in Offline Payments with Reputation-Weighted Loan Networks

Blockchain solutions typically assume a synchronous network to ensure consistency and achieve consensus. In contrast, offline transaction systems aim to enable users to agree on and execute transactions without assuming bounded communication delays when interacting with the blockchain. Most existing offline payment schemes depend on trusted hardware wallets that are assumed to be secure and tamper-proof. While this work introduces Overdraft, a novel offline payment system that shifts the reliance from hardware to users themselves. Overdraft allows potential payment receivers to assess the likelihood of being paid, allowing them to accept transactions with confidence or deny them. Overdraft achieves this by maintaining a loan network that is weighted by online reputation. This loan network contains time-limited agreements where users pledge to cover another user's payment if necessary. For example, when a payer lacks sufficient funds at the moment of commitment. Offline users rely on the last known view of the loan network -- which they had access to when last online -- to determine whether to participate in an offline transaction. This view is used to estimate the probability of eventual payment, possibly using multiple loans. Once online again, users commit their transactions to the blockchain with any conflicts being resolved deterministically. Overdraft incorporates incentives for users and is designed to be resilient against Sybil attacks. As a proof of concept, we implemented Overdraft as an Ethereum Solidity smart contract and deployed it on the Sepolia testnet to evaluate its performance.

cs.CR↗

arXiv2025

SoK: Microservice Architectures from a Dependability Perspective

The microservice software architecture leverages the idea of splitting large monolithic applications into multiple smaller services that interact using lightweight communication schemes. While the microservice architecture has proven its ability to support modern business applications, it also introduces new possible weak points in a system. Some scientific literature surveys have already addressed fault tolerance or security concerns but most of them lack analysis on the fault and vulnerability coverage that is introduced by microservice architectures. We explore the known faults and vulnerabilities that microservice architecture might suffer from, and the recent scientific literature that addresses them. We emphasize runtime detection and recovery mechanisms instead of offline prevention and mitigation mechanisms to limit the scope of this document.

cs.DC↗

arXiv2024

Optimizing Streamlined Blockchain Consensus with Generalized Weighted Voting and Enhanced Leader Rotation

Streamlined Byzantine Fault Tolerant (BFT) protocols, such as HotStuff [PODC'19], and weighted voting represent two possible strategies to improve consensus in the distributed systems world. Several studies have been conducted on both techniques, but the research on combining the two is scarce. To cover this knowledge gap, we introduce a weighted voting approach on Hotstuff, along with two optimisations targeting weight assignment distribution and leader rotation in the underlying state replication protocol. Moreover, the weighted protocols developed rely on studies proving the effectiveness of a specific voting power assignment based on discrete values. We generalise this approach by presenting a novel continuous weighting scheme applied to the Hotstuff protocol to highlight the effectiveness of this technique in faulty scenarios. We prove the significant latency reduction impact of weighted voting on streamlined protocols and advocate for further research.

cs.DC↗