HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Diffusion-based Frameworks for Unsupervised Speech Enhancement
No full textThis paper addresses unsupervised diffusion-based single-channel speech enhancement (SE). Prior work in this direction combines a score-based diffusion model trained on clean speech with a Gaussian noise model whose covariance is structured by non-negative matrix factorization (NMF). This combination is used within an iterative expectation–maximization (EM) scheme, in which a diffusion-based posterior-sampling E-step estimates the clean speech. We first revisit this framework and propose to explicitly model both speech and acoustic noise as latent variables, jointly sampling them in the E-step instead of sampling speech alone as in previous approaches. We then introduce a new unsupervised SE framework that replaces the NMF noise prior with a diffusion-based noise model, learned jointly with the speech prior in a single conditional score model. Within this framework, we derive two variants: one that implicitly accounts for noise and one that explicitly treats noise as a latent variable. Experiments on WSJ0–QUT and VoiceBank–DEMAND show that explicit noise modeling systematically improves SE performance for both NMF-based and diffusion-based noise priors. Under matched conditions, the diffusion-based noise model attains the best overall quality and intelligibility among unsupervised methods, while under mismatched conditions the proposed NMF-based explicit-noise framework is more robust and suffers less degradation than several supervised baselines
Assessing Graph Neural Networks for latency and power consumption prediction in application mappings on multicore architectures
No full textInternational audienceAccurately estimating the latency and power consumptionof software applications deployed on multicore systems remains a majorchallenge for early-stage optimization, as existing methods typically relyon slow and resource-intensive simulations. This paper explores modelingapplication-to-architecture mappings as heterogeneous graphs and investigatesGraph Neural Networks (GNNs) for predicting their performance.Four GNN models are evaluated across eleven datasets, considering fiveNeural Network-based software applications. The two best models achievemean absolute percentage errors of about 2% for power prediction and 15%for latency, with prediction times of only a few tens of milliseconds. Theseresults indicate the potential of GNN-based prediction as an efficient alternativeto simulation-driven estimation, paving the way for early-stageAI-assisted mapping optimization
Air Gap Shape Optimization for Minimizing Proximity Losses in an Inductor
No full textInternational audienceIn air-gapped inductors, the magnetic flux bulges outward from the gap -a phenomenon called the fringing effect. When these flux lines cut across the conductor windings, they generate undesired eddy currents that lead to substantial AC losses and reduced efficiency. To mitigate these losses, we develop a constrained shape optimization, aiming at finding an optimal air gap profile that minimizes conductor losses and keeps the inductance at a desired value. Our framework leverages established techniques such as complex permeability homogenization for loss modeling, shape sensitivity analysis coupling the adjoint method with the so-called Hilbertian regularization-extension method. The optimized air gap profile is curved to concentrate the flux lines away from the (fixed) conductors. Our implementation uses the open-source finite-element toolbox NGSolve and the Mmg remeshing library; all the codes are freely available.</div
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
Live-cell SICM imaging: An Introductory Guide for New Users
No full textHigh-resolution, minimally invasive imaging of live cells is essential for investigating cellular morphology and its dynamic changes. Among available approaches, Scanning Ion Conductance Microscopy (SICM) offers a unique combination of precise topographic imaging down to nanoscale with experimental conditions that preserve true livecell behavior. Here, we present a practical guide based on our own experience and experiments, intended to be a resource to help first-time SICM users, covering critical aspects from instrumentation, probe characterization, to cell preparation and morphometric data extraction and presentation, to accelerate their implementation, learning and to gain confidence in exploring live-cell structure and dynamics
Multi-period replenishment planning with supplier assignment under a dynamic demand and stochastic lead-times
No full textInternational audienceWe consider a multi-period replenishment planning problem with dynamic demand and multiple suppliers. Each supplier has its own selling price and random lead time. The objective is to decide how the demands for different periods are distributed among the pre-selected suppliers while minimising the expected total cost, which comprises holding and backlogging costs, as well as the selling price. This statement highlights the trade-off between supplier prices and uncertain lead times in replenishment planning. Two scenario-based stochastic programmes, one linear and one non-linear, are provided to simultaneously consider dispatching orders between suppliers, order crossover, and order release flexibility. The non-linear model, which relies on power sets to reduce the number of aggregated scenarios, is coupled with two approximate solution methods. The numerical experiments prove the computational effectiveness of the non-linear model. These results can be helpful for decision-makers when negotiating prices with suppliers. For instance, we can determine the purchasing cost at which we order almost 100% from a given supplier. If we now turn to the uncertainty of lead times, this study can also help negotiate with suppliers whose firms cannot decrease their selling prices and advise them, thereby reducing the variance of their lead times
Quantum Coherence Spaces Revisited: A von Neumann (Co)Algebraic Approach
No full textInternational audienceWe describe a categorical model of MALL (Multiplicative Additive Linear Logic) inspired by the Heisenberg-Schrödinger duality of finite-dimensional quantum theory. Proofs of formulas with positive logical polarity correspond to CPTP (completely positive trace-preserving) maps in our model, i.e. the quantum operations in the Schrödinger picture, whereas proofs of formulas with negative logical polarity correspond to CPU (completely positive unital) maps, i.e. the quantum operations in the Heisenberg picture. The mathematical development is based on noncommutative geometry and finite-dimensional von Neumann (co)algebras, which can be defined as special kinds of (co)monoid objects internal to the category of finite-dimensional operator spaces
Evaluating the strength of molecular interactions in a deep eutectic solvent (DES) by means of ionization mechanisms involved in cold-spray ionization mass spectrometry and by DFT calculations
No full textInternational audienceStrength of molecular interactions in octanoic acid and cholinium chloride based DESs determined using CSI-MS and DFT calculations
Certifying the Decidability of the Word Problem in Monoids at Large
No full textInternational audienceWhile the word problem for monoids is undecidable in general, having a decision procedure for some finitely presented monoid of interest has numerous applications. This paper presents a toolbox for the Rocq proof assistant that can be used to verify the decidability of the word problem for a given monoid and, in some cases, to produce the corresponding decision procedure. As this verification can be computationally intensive, the toolbox heavily relies on proofs by reflection guided by an external oracle. This approach has been successfully used on several large presentations from the literature, as well as on a database of one million 1-relation monoids.The huge size of this database forced some unusual considerations onto the Rocq formalization, so that the formal proofs could be checked in a reasonable amount of time.</p
Model Variability in Assessment of Human Exposure to Radiofrequency Fields
No full textInternational audienceThe recent advances in computational dosimetry for electromagnetics and thermodynamics are reviewed to assess human exposure to electromagnetic fields in the MHz-to-terahertz range. This review emphasizes model variability in computational dosimetry. Apart from computational electromagnetic methods and their usage, the developments in anatomical phantoms and tissue dielectric properties characterization are also surveyed. In addition, the rationale for dosimetric quantities prescribed in international exposure guidelines, such as the specific absorption rate (SAR) and absorbed power density, is revisited in relation to their correlation with local and core temperature rises in various tissues and populations. A heating factor, which is defined as a steady-state temperature rise per SAR, for the brain, eye lens, skin, and body core is evaluated to estimate heating resulting from exposure to electromagnetic fields. The transition of a physical quantity in the guidelines at 6 GHz, from SAR to the absorbed power density, is discussed along with the optimal spatial averaging volume and areas. Computational evaluations of product compliance, 5G devices, and wireless power transfer systems are also reviewed. This review aims to synthesize the current knowledge, identify key sources of computational model variability and uncertainty, and outline further research needs for setting exposure guidelines and compliance assessment