HAL-Ecole des Ponts ParisTech
Not a member yet
42743 research outputs found
Sort by
Sliced ReLU attention: Quasi-linear contextual expressivity via sorting
No full textWe introduce sliced ReLU attention, a new attention mechanism that departs structurally from both softmax and ReLU-based alternatives. Instead of applying a nonlinearity to pairwise dot products, we operate on one-dimensional projections of key–query differences and leverage sorting to obtain quasi-linear complexity. This construction yields a differentiable, non-symmetric kernel that can be computed in O(n log(n)) through a sorting procedure, making it suitable for very long contexts. Beyond computational benefits, the model retains strong theoretical expressive power: we establish two in-context expressivity results, previously known for softmax attention, showing that sliced ReLU attention preserves the ability to perform nontrivial sequence-to-sequence disentangling tasks and satisfies a contextual universal approximation property. Finally, we illustrate the potential practical interest of this kernel in small-scale experiments
Low-cost electrochemical gas sensing of vertical differences in wintertime air composition (CO, NO, NO2, O3) in Fairbanks, Alaska
No full textInternational audienceWintertime Fairbanks, Alaska, experiences episodes of severely poor air quality, when local emissions (e.g., home-heating, vehicular) are enhanced by cold conditions and are trapped by temperature inversions. Monitoring of atmospheric composition, and in particular vertical gradients in composition, is challenging under cold Arctic conditions. This study demonstrates that multiple sets of low-cost electrochemical sensors can provide accurate measurement of CO, NO, NO2, and O3 air composition across wide-ranging cold Arctic temperatures (0 °C to −30 °C). The sensors quantify vertical gradients in downtown Fairbanks’ atmospheric composition during winter 2021. Low-cost electrochemical sensors (with temperature co-measured) were characterised by cross-comparison to a regulatory air-quality monitoring site. We demonstrate excellent agreement of the electrochemical sensors with the reference monitors (R2 > 0.86–0.98), with mean absolute errors <5 ppbv (NO, NO2, O3) and <50 ppbv (CO) over gas-ranges of 10–100’s, and 3000 ppbv, respectively, sufficient for using the low-cost electrochemical sensors to quantitatively investigate NO–NO2–O3 atmospheric chemistry. During four weeks in February–March 2021, sensors placed on the rooftop (20 m) and base (3 m) of a building in downtown Fairbanks identified strong gradients in atmospheric composition over a very short <20 m vertical scale at times when near-surface temperature inversions were present. At night, CO and NOx were more concentrated at the surface than aloft, and surface ozone was depleted whilst sometimes being present aloft. During the daytime, when solar radiation heated the surface, inversions were disrupted by efficient vertical mixing that mixed in ozone-rich air from above. The low-cost sensor observations demonstrate that near-surface pollutant trapping was correlated with thermal inversions and trace O3–NOx atmospheric chemistry, and quantify a local Ox source from direct “primary” NO2 emissions, with a directly emitted NO2 : NOx ratio of 0.13 mol mol−1. The sensors also characterise NOx emissions, finding a NOx : CO of 0.15 mol mol−1. When well-characterised, low-cost electrochemical sensors can provide valuable measurements of local emissions and vertically-resolved atmospheric composition, with sufficient accuracy to trace atmospheric chemistry in cold and stable wintertime urban environments
Database on fungal OTUs (ITS1) on graves in Père Lachaise Cemetery, Paris, France
No full textInternational audienceThis database includes all FASTQ files for fungal OTUs ITS 1 collected from graves in Père Lachaise Cemetery and a txt file explaining the date and method of sampling, the stone type of the grave and the GPS coordinates. You can find more details in Paloma Reboah's thesis. https://hal.science/tel-04777374 ITS15':CCTACGGGNGGCWGCAG3':AAACTYAAAKRAATTGRCG
Morse sequences on stacks and flooding sequences
No full textThis paper builds upon the framework of \emph{Morse sequences}, a simple and effective approach to discrete Morse theory. A Morse sequence on a simplicial complex consists of a sequence of nested subcomplexes generated by expansions and fillings—two operations originally introduced by Whitehead. Expansions preserve homotopy, while fillings introduce critical simplexes that capture essential topological features.We extend the notion of Morse sequences to \emph{stacks}, which are monotonic functions defined on simplicial complexes,and define \emph{Morse sequences on stacks} as those whose expansions preserve the homotopy of all sublevel sets. This extension leads to a generalization of the fundamental collapse theorem to weighted simplicial complexes. Within this framework, we focus on a refined class of sequences called \emph{flooding sequences}, which exhibit an ordering behavior similar to that of classical watershed algorithms. Although not every Morse sequence on a stack is a flooding sequence, we show that the gradient vector field associated with any Morse sequence can be recovered through a flooding sequence.Finally, we present algorithmic schemes for computing flooding sequences using cosimplicial complexes
Search Schemes for Approximate Pattern Matching: An Overview
No full textInternational audienceWe provide a brief survey of results on solving the approximate pattern matching problem using search schemes, as introduced by Kucherov et al. (2016). We demonstrate that search schemes constitute a flexible and versatile tool that enable the specification of various search strategies, including several known filtering methods. We present approaches for designing efficient search schemes and for implementing them effectively. Finally, we conclude with experimental results comparing multiple search schemes on DNA sequencing data using the Columba software by Renders et al. (2021)
Linking European Temperature Variations to Atmospheric Circulation With a Neural Network: A Pilot Study in a Climate Model
No full textInternational audienceIn Europe, temperature variations are mainly driven by the North Atlantic atmospheric circulation. Here, with data from the MIROC6 large ensemble, we investigate a convolutional neural network (a UNET) for reconstructing daily temperature anomalies in Europe from Sea Level Pressure (SLP) as a proxy of the atmospheric circulation, and we compare the results with a traditional analogs approach. We show an excellent ability of the UNET to estimate temperature variations given information from SLP only. This novel method outperforms the analogs method, at both daily and inter‐annual time scales. Our study also shows that during the training, the UNET learns information such as the seasonal cycle of the relationship between sea‐level pressure and temperature anomalies, which could explain part of its excellent scores. This exploratory work opens up promising prospects for estimating the contribution of atmospheric variability to observed temperature variations
Towards diffusion models for large-scale sea-ice modelling
No full textWe make the first steps towards diffusion models for unconditional generation of multivariate and Arctic-wide sea-ice states. While targeting to reduce the computational costs by diffusion in latent space, latent diffusion models also offer the possibility to integrate physical knowledge into the generation process. We tailor latent diffusion models to sea-ice physics with a censored Gaussian distribution in data space to generate data that follows the physical bounds of the modelled variables. Our latent diffusion models reach similar scores as the diffusion model trained in data space, but they smooth the generated fields as caused by the latent mapping. While enforcing physical bounds cannot reduce the smoothing, it improves the representation of the marginal ice zone. Therefore, for large-scale Earth system modelling, latent diffusion models can have many advantages compared to diffusion in data space if the significant barrier of smoothing can be resolved
Suggestions for unifying the parameterizations that represent turbulent orographic form drag, mountain wave drag and flow blocking
No full textInternational audienceParameterizations of subgrid scale mountains are commonly used in numerical weather prediction and climate models. They try to represent quite separate processes namely the enhancement of the turbulent drag by orography, gravity wave drag, and the effects of low level flow blocking. Among the gravity wave schemes, some of them distinguish between upward propagating waves and trapped lee waves. This paper makes use of a recent theoretical methodology to propose a formalism that include all these effects. This theory handles enhanced turbulent drag in the neutral case, gravity waves in the stratified case, and trapped lee waves in the transition. Mountain drag associated to all these processes is estimated analytically, as well as the fraction of the drag that stays within the boundary layer instead of being radiated in the far field. We also try to evaluate the blocked layer depth by combining the sheltering effects that dominate when stratification is small and the blocking effects when stratification is strong
Penser le temps de la transition. La dimension socio-technique de la ville-territoire
No full textInternational audienceInterroger sur la longue durée, la question des transitions socio-techniques, permet d’observer que les techniques nouvelles ou renaissantes se sont plus agrégées que substituées aux précédentes, formant des réseaux hybrides, consolidant des macro-systèmes techniques. En définitive, les intentions qui sont à l’origine de leur introduction ne sont pas acquises, parfois même contrariées suivant les effets rebonds ou les contre-productivités. Autrement dit, loin d’opérer des ruptures dans les systèmes de consommation et de production, les transitions peinent à nous émanciper des trajectoires socio-techniques et de leurs verrouillages. À travers la question du transport et des cultures mobiles, nous nous intéresserons à ce nexus technique/politique et à ses conséquences sur la transition socio-écologique qui sont au cœur de l’agir public, et que mettent en lumière la montée des controverses et la difficulté de mettre en place des instruments efficaces pour gouverner la transition. Dans cette intervention, nous nous attacherons, à travers les cas des villes du Grand Paris, à comprendre les trajectoires qui façonnent nos environnements urbains et les possibilités de bifurquer. Situer l’action dans les villes et les territoires, permet de clarifier ce que font les processus d’aménagement à l’espace. En posant la question de l’infrastructure comme relation ; en interrogeant la relation de l'espace à la vitesse ; en pensant par les modes de vie; en travaillant la transversalité des cultures professionnelles et des arènes démocratiques
