INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Making Democracy Work: Fixing and Simplifying Egalitarian Paxos
No full textInternational audienceClassical state-machine replication protocols, such as Paxos, rely on a distinguished leader process to order commands. Unfortunately, this approach makes the leader a single point of failure and increases the latency for clients that are not co-located with it. As a response to these drawbacks, Egalitarian Paxos [Iulian Moraru et al., 2013] introduced an alternative, leaderless approach, that allows replicas to order commands collaboratively. Not relying on a single leader allows the protocol to maintain non-zero throughput with up to f crashes of any processes out of a total of n = 2f+1. The protocol furthermore allows any process to execute a command c fast, in 2 message delays, provided no more than e = ⌈(f+1)/2⌉ other processes fail, and all concurrently submitted commands commute with c; the latter condition is often satisfied in practical systems. Egalitarian Paxos has served as a foundation for many other replication protocols. But unfortunately, the protocol is very complex, ambiguously specified and suffers from nontrivial bugs. In this paper, we present EPaxos* - a simpler and correct variant of Egalitarian Paxos. Our key technical contribution is a simpler failure-recovery algorithm, which we have rigorously proved correct. Our protocol also generalizes Egalitarian Paxos to cover the whole spectrum of failure thresholds f and e such that n ≥ max{2e+f-1, 2f+1} - the number of processes that we show to be optimal
Assessing Stimuli Detectability and Pleasantness for Auditory BCI
No full textInternational audienceBrain-Computer Interfaces (BCIs) enable device control by analyzing brain activity. In reactive auditory BCIs based on steady-state auditory evoked potentials, users are exposed to amplitude-modulated sine waves at given frequencies that encode information (i.e. the type of action expected), while their brain activity is analyzed to infer the intended action based on the frequency retrieved. However, listening to sine-wave may be perceived as unpleasant over time. This study compares the use of pure-tones with alternative sounds, including artificial stimuli (such as a Brownian noise) and natural sounds (such as cicada song and cat's purr) by measuring brain responses of 48 subjects to these different stimuli, all amplitude-modulated at 40 Hz. The Signal-to-Noise Ratio (SNR) (i.e. the ratio between the power spectrum of electroencephalographic signals in response to the target stimulus and that in response to a silence stimulus) is computed at 40 Hz for each type of stimulus. It reveals that the 40-Hz modulation frequency is clearly more identifiable when carried by a pure tone than when carried by the other sounds, with an SNR increase up to more than 5 dB. The cicada song stimulus is a promising alternative, still requiring improvement to achieve the level of detectability observed for pure tones. The experiment is conducted at two different sound levels to assess whether increasing the listening level increases the SNR, but the opposite trend is found. Questionnaires indicate that more than half of the participants find pure tones annoying and prefer other sounds, confirming that this study is worth pursuing
Frontiers of Combining Systems. 15th International Symposium, FroCoS 2025, Reykjavik, Iceland, September 29 – October 1, 2025, Proceedings
No full textInternational audienc
Mathematical Modeling of Mucus Transport in the Bronchial Tree with Ventilation Effects
No full textMucociliary clearance plays a central role in respiratory defense by transporting mucus along the bronchial tree through the combined action of cilia, mucus secretion, and airflow. We propose a multiscale mathematical model of this process that explicitly couples ciliary-driven transport, mucus production, and ventilation effects. The model combines a one-dimensional variational formulation for mucus velocity within individual airways with a system of ordinary differential equations describing the evolution of mucus thickness across bronchial generations. Realistic airway geometry and breathing scenarios are incorporated. The analysis reveals stable equilibrium mucus distributions in the absence of ventilation and periodic regimes under physiological breathing. Numerical results quantify the impact of secretion, ciliary activity, and airflow, and highlight nonlinear feedbacks between mucus accumulation and airway resistance, with potential relevance for pathological conditions
BEST-RQ-Based Self-Supervised Learning for Whisper Domain Adaptation
No full textInternational audienceAutomatic Speech Recognition (ASR) systems, despite large multilingual training, struggle in low-resource scenarios where labeled data is scarce. We propose BEARD (BEST-RQ Encoder Adaptation with Re-training and Distillation), a novel framework designed to adapt Whisper’s encoder with unlabeled data. Unlike traditional self-supervised learning methods, BEARD uniquely combines a BEST-RQ objective with knowledge distillation from a frozen teacher encoder, ensuring the encoder's complementarity with the pre-trained decoder. Our experiments focus on the ATCO2 corpus from the challenging Air Traffic Control (ATC) communications domain, characterized by non-native speech, noise, and specialized phraseology. Using about 5,000 hours of untranscribed speech for BEARD and 2 hours of transcribed speech for fine-tuning, the proposed approach significantly outperforms previous baseline and fine-tuned model, achieving a relative improvement of 12% compared to the fine-tuned model. To the best of our knowledge, this is the first work to use a self-supervised learning objective for domain adaptation of Whisper
Intrinsic Estimation of Frenet Curvatures for 3D Trajectory Analysis
No full textInternational audienceThis paper addresses a statistical challenge underlying the precise analysis of curves in from discrete and noisy observations. A significant part of curve or functional data analysis relies heavily on understanding the shapes of curves. While Frenet curvature parameters are recognized for their ability to capture the full geometric characteristics of curves, their application in functional data analysis has been limited due to the complexities of their statistical estimation. We address this gap by providing a comprehensive review of existing methods for estimating Frenet curvatures, which often depend on latent variable estimation during preprocessing. Additionally, we introduce a new unified intrinsic approach that leverages the Lie group SE(3) to offer a more robust and reliable estimation of Frenet parameters, aiming to advance the precision of shape-based curve analysis
Towards a Novel Vertical Scaling Approach for Bursty Workloads in Kubernetes
No full textInternational audienceTraditional static computational resource allocation in cloud oron-premises clusters often results in inefficient overprovisioning.Users frequently lack precise knowledge of the memory and processors their applications require, leading them to request excessresources. This causes wasted capacity, higher costs, and, in sharedenvironments, longer queue waiting times. Dynamic resource allocation through autoscaling addresses this issue by adjusting resources at runtime. Kubernetes, a widely used container orchestration platform, supports autoscaling via Horizontal and Vertical PodAutoscalers. However, its default restart-based scaling can disruptstateful, long-running workloads without checkpointing. This workleverages Kubernetes’ new in-place scaling, which resizes resourceswithout restarts, to propose the Dynamic Resizing Strategy (DRS), anovel autoscaling approach that proactively manages contention bytemporarily throttling co-located pods to prioritize a bursting application. We evaluate it with NAS Parallel Benchmarks and syntheticworkloads in co-execution scenarios, showing improved efficiencyand stability, increasing success rates and reducing global averagewait time by over 18% compared to the Burstable QoS class
Reducing the Hardware Gap for Custom Accelerators through Quantization Aware Training
No full textInternational audiencePopular machine learning frameworks like PyTorch and TensorFlow provide complete toolchains to design, train, quantize and deploy convolutional neural networks (CNNs) on standard hardware (CPUs, GPUs, TPUs, NPUs, microcontrollers). However, custom hardware targets like FPGAs, ASICs and research accelerators typically use non-standard number representations as well as limited operator sets, and are poorly served by current lowering backends. Existing backend-driven approaches rigidly force training to their internal formats and often hide lowering choices behind opaque layers and closed software development kits. Conversely, unconstrained quantization that permits arbitrary numeric and operator freedom often produces models that are difficult to implement efficiently, leading to an hardware gap between the trained and deployed model. This work introduces HATorch, a PyTorch-based hardwareaware training framework that reduces this gap by enhancing quantization with hardware architecture model on the arithmetic operator/format level. HATorch supports custom hardware-friendly quantization flows, exposing lowering decisions for transparent model-hardware co-design
Folded optimal transport and its application to separable quantum optimal transport
No full textWe introduce folded optimal transport, as a method to extend a cost or distance defined on the extreme boundary of a convex to the whole convex, related to convex extension. This construction broadens the framework of standard optimal transport, found to be the particular case of the convex being a simplex. Relying on Choquet's theory and standard optimal transport, we introduce the folded Kantorovich cost and folded Wasserstein distances, and study their induced metric properties. We then apply the construction to the quantum setting, and obtain an actual separable quantum Wasserstein distance on the set of density matrices from a distance on the set of pure states, closely related to the semi-distance of Beatty and Stilck-França (2025), and of which we obtain a variety of properties. We also find that the semiclassical Golse-Paul (2017) cost writes as a folded Kantorovich cost. Folded optimal transport therefore provides a unified framework for classical, semiclassical and separable quantum optimal transport
Diagrammatic Reasoning with Control as a Constructor, Applications to Quantum Circuits
No full textInternational audienceControl is a fundamental concept in quantum and reversible computational models. It enables the conditional application of a transformation to a system, depending on the state of another system. We introduce a general framework for diagrammatic reasoning featuring control as a constructor. To this end, we provide an elementary axiomatisation of control functors, extending the standard formalism of props to controlled props. As an application, we show that controlled props facilitate diagrammatic reasoning for quantum circuits by introducing a simple and complete set of relations involving at most three qubits, whereas in the standard prop setting any complete axiomatisation necessarily requires relations acting on arbitrarily many qubits