HAL-Polytechnique
Not a member yet
51406 research outputs found
Sort by
Is Contextual Advertising Safe? Analyzing Systemic Risks with Ads on YouTube
No full textInternational audienceContextual advertising is seeing a resurgence in popularity as a privacy-preserving alternative to behavioral advertising. While often regarded as a coarse-grained approach, advances in AI-driven content analysis have transformed it into a highly granular form of targeting. In this paper, we examine the safety risks of contextual advertising through a two-part empirical study of YouTube. First, we show that advertisers can reach audiences defined by sensitive attributes-such as religion, mental health conditions, and political ideology-by strategically selecting contextual placements, effectively circumventing policies that prohibit such targeting via behavioral signals. Second, to assess how this risk manifests in practice, we develop an automated measurement framework to collect ads delivered on high-risk content environments, focusing on conspiracy videos. We find that contextual advertising is highly prevalent in these environments, disproportionately delivers sensitive ad categories (including alternative health, religion, and political content), and provides limited transparency into why ads are shown. Our results demonstrate that contextual advertising can be as opaque and susceptible to abuse as behavioral targeting, and highlight the need for stronger transparency and regulatory scrutiny of contextual ad systems
Water vapour isotope anomalies during an atmospheric river event at Dome C, East Antarctica
No full textInternational audienceAbstract. From 19 to 23 December 2018, an atmospheric river sourced in the Atlantic hit the French–Italian Concordia station, located at Dome C, East Antarctic Plateau, 3233 m above sea level (a.s.l.). It induced a significant surface warming (+18 °C in 3 d), combined with high specific humidity (3-fold increase in 3 d) and a strong isotopic anomaly in water vapour (+17 ‰ for δ18O). The isotopic composition of water vapour monitored during the event may be explained by the isotopic signature of long-range water transport, and by local moisture uptake during the event. In this study, we used continuous meteorological and isotopic water vapour observations, together with the atmospheric general circulation model LMDZ6iso, to describe this event and quantify the influence of each of these processes. The presence of mixed-phase clouds during the event induced a significant increase in downward long wave radiation, leading to high surface temperature and resulting in high turbulent mixing in the boundary layer. Although surface fluxes are underestimated in LMDZ6iso, near-surface temperature and specific humidity are well represented. The surface vapour δ18O is accurately simulated during the event, despite an overestimated amplitude in the diurnal cycle outside of the event. Using the LMDZ6iso simulation, we perform a surface water vapour mass budget by decomposing total specific humidity into contributions from individual processes. Our analysis demonstrates that surface sublimation, which becomes significantly stronger during the event compared to typical diurnal cycles, emerges as the dominant driver of the vapour δ18O signal at the peak of the event, accounting for approximately 70 % of the total contribution. The second largest contribution comes from moisture input via large-scale advection associated with the atmospheric river, accounting for approximately 30 % of the total. Consequently, our results reveal that the isotopic signal monitored in water vapour during this atmospheric river event reflects both long-range moisture advection and interactions between the boundary layer and the snowpack. Only specific meteorological conditions driven by a pronounced moisture intrusion can explain these strong interactions. Given the marked imprint of air–snow exchanges on the vapour isotopic signal, improving the representation of local processes in climate models could substantially improve the simulation of the isotopic signal over Antarctica and provide valuable insight into moisture uptake processes
Empirical distribution of ancestral lineages in populations with density-dependent interactions
No full textWe study a density-dependent Markov jump process describing a population where each individual is characterized by a type, and reproduces at rates depending both on its type and on the population type distribution. We are interested in the empirical distribution of ancestral lineages in the population process. First, we exhibit a time-inhomogeneous Markov process, which allows to capture the behavior of a sampled lineage in the population process. This is achieved through a many-to-one formula, which relates the expected value of a functional evaluated over the lineages in the population process to the expectation of the functional evaluated along this time-inhomogeneous process. This provides a direct interpretation of the underlying survivorship bias, as illustrated on a minimalistic population process. Second, we consider the large population regime, when the population size grows to infinity. Under classical assumptions, the population type distribution converges to a deterministic limit.Here, we focus on the empirical distribution of ancestral lineages in this large population limit, for which we establish a many-to-one formula. Using coupling arguments, we further quantify the approximation error which arises when sampling in this large population approximation instead of the finite-size population process
First results from the E302 efficiency\unicode{x2013}instability experiment at the FACET-II facility
No full textInternational audienceThe beam-breakup (BBU) instability in plasma accelerators is seeded by a transverse offset between the driver and a trailing bunch. The BBU instability induces oscillations in the trailing bunch, which are detrimental to its beam quality. When the instability is large, assuming little mitigation from ion motion and energy spread, the beam suffers emittance growth, and charge can be kicked transversely out of the plasma channel. The detrimental effect on beam quality is substantially worse at high efficiencies, which places constraints on the achievable power efficiency in applications such as linear colliders, where maintaining the beam quality is required. In this paper, we present the first experimental signatures of the BBU instability in data taken in the E302 experiment at the FACET-II facility at SLAC National Accelerator Laboratory. We use a specific beam-optical setup and a novel method to probe for transverse instabilities on diagnostic screens downstream of a magnetic dipole spectrometer. We complement the analysis with full 3D particle-in-cell (PIC) simulations of the plasma interaction using similar driver and trailing bunch parameters on a simulated FACET-II spectrometer
Proving symmetry of localized solutions and application to dihedral patterns in the planar Swift-Hohenberg PDE
No full textIn this article, we extend the framework developed in [14] to allow for rigorous proofs of existence of smooth, localized solutions in semi-linear partial differential equations possessing both space and non-space group symmetries. We demonstrate our approach on the Swift-Hohenberg model. In particular, for a given symmetry group G, we construct a natural Hilbert space H l G containing only functions with G-symmetry. In this space, products and differential operators are well-defined allowing for the study of autonomous semi-linear PDEs. Depending on the properties of G, we derive a Newton-Kantorovich approach based on the construction of an approximate inverse around an approximate solution, u0. More specifically, combining a meticulous analysis and computer-assisted techniques, the Newton-Kantorovich approach is validated thanks to the computation of some explicit bounds. The strategy for constructing u0, the approximate inverse, and the computation of these bounds will depend on the properties of G and its maximal square lattice space subgroup, H. More specifically, we consider three cases: G is a space group which can be represented on the square lattice, G is not a space group which can be represented on the square lattice and the symmetry of H isolates the solution, and where G is not a space group which can be represented on the square lattice and the symmetry of H does not isolate the solution. We demonstrate the methodology on the 2D Swift-Hohenberg PDE by proving the existence of various dihedral localized patterns. The algorithmic details to perform the computer-assisted proofs can be found on Github [4].</div
Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings
No full textInternational audienceOne-step generators distilled from Masked Diffusion Models (MDMs) compress multiple sampling steps into a single forward pass, enabling efficient text and image synthesis. However, they suffer two key limitations: they inherit modeling bias from the teacher, and their discrete token outputs block gradient flow, preventing post-distillation refinements such as adversarial training, reward-based fine-tuning, and Test-Time Embedding Optimization (TTEO). In this work, we introduce soft embeddings, a simple relaxation that replaces discrete tokens with the expected embeddings under the generator's output distribution. Soft embeddings preserve representation fidelity for one-step discrete generator while providing a fully differentiable continuous surrogate that is compatible with teacher backbones and tokenizer decoders. Integrating soft embeddings into the Di[M]O distillation framework (denoted Soft-Di[M]O) makes one-step generators end-to-end trainable and enables straightforward application of GAN-based refinement, differentiable reward fine-tuning, and TTEO. Empirically, across multiple MDM teachers (e.g., MaskBit, MaskGen), Soft-Di[M]O achieves state-of-the-art one-step results: improved class-to-image performance, a one-step FID of 1.56 on ImageNet-256 with GAN-based refinement, along with higher GenEval and HPS scores on text-to-image with reward fine-tuning, and further gains from TTEO
Disturbed and quiet days ∑O/N<sub>2</sub> variations at low and mid-latitudes during solar cycles 23 and 24
No full textInternational audienceWe analyzed the column density ratio of thermospheric compositions (∑O/N2) using data from the Global Ultraviolet Imager (GUVI) onboard the TIMED satellite from 2002 to 2020. Daily ∑O/N 2 values for the three most geomagnetically disturbed and quietest days each month were used to compute monthly means at low and mid-latitudes across both hemispheres. These variations were also examined across various longitudinal sectors, including Asia, Africa, and the Americas. The fluctuations in ∑O/N2 were more pronounced at mid-latitudes than at low latitudes, with low latitude values in both hemispheres peaking near the equinoxes. At midlatitudes, the highest values occurred during local winter, with stronger peaks in the Northern Hemisphere (NH) than in the Southern Hemisphere (SH). The winter and equinoctial maxima are also observed in all longitudinal sectors. Besides this, the distinct longitudinal asymmetries over Asian, African, and American regions at mid-latitudes, influenced by geomagnetic field geometry, are also observed. The downwelling of ∑O/N2 in local winter is stronger, while upwelling in local summer is weaker in the longitudinal sectors containing the magnetic pole. Annual (AV) and semiannual variations (SAV) were extracted using a bandpass filter. AV was stronger at mid-latitudes, peaking in local winter and highlighting the winter anomaly in both hemispheres. SAV were dominant at low latitudes, with positive peaks at equinoxes and negative dips at solstices, generally in phase across hemispheres and longitude sectors. The amplitudes of AV and SAV are stronger during solar maximum periods, justifying the solar cycle trend. Analysis also revealed that during geomagnetically disturbed periods, ∑O/N2 typically decreased (≤ -10%) at mid-latitudes and increased (≥10%) at low latitudes compared to quiet periods. Although opposite trends-enhancement at mid-latitudes and depletion at low latitudes -were occasionally observed, they were less significant. This study aims to provide valuable insights into the dynamics of thermospheric composition, thereby contributing to the improved modeling of ionospheric behavior and space weather forecasting
TRAKNN-Explorer: Interactive Discovery of Rare Meteorological Trajectories and Analogues Retrieval
No full textExtreme weather events often arise from multi-day atmospheric evolutions that cannot be captured by traditional snapshot-based similarity methods. We present TRAKNN-Explorer, an interactive system for discovering rare meteorological trajectories and retrieving trajectorybased analogues in spatiotemporal datasets. The interface allows users to configure trajectory parameters, explore rarity scores, visualize spatial anomalies, cluster rare events, and retrieve historical trajectory analogues. During the demonstration, participants will interact with the full pipeline using ERA5 reanalysis data to explore rare weather evolutions
Asymptotics of SGD in Sequence-Single Index Models and Single-Layer Attention Networks
No full textWe study the dynamics of stochastic gradient descent (SGD) for a class of sequence models termed Sequence Single-Index (SSI) models, where the target depends on a single direction in input space applied to a sequence of tokens. This setting generalizes classical single-index models to the sequential domain, encompassing simplified one-layer attention architectures. We derive a closed-form expression for the population loss in terms of a pair of sufficient statistics capturing semantic and positional alignment, and characterize the induced high-dimensional SGD dynamics for these coordinates. Our analysis reveals two distinct training phases: escape from uninformative initialization and alignment with the target subspace, and demonstrates how the sequence length and positional encoding influence convergence speed and learning trajectories. These results provide a rigorous and interpretable foundation for understanding how sequential structure in data can be beneficial for learning with attention-based models
Effect of polyelectrolyte mixing ratio and hydrophobic interactions on dynamics of (HM-)PDMAEMA/PEO-PMAA complexes
No full textInternational audienceThe complexation of oppositely charged polyelectrolytes leads to Polyelectrolyte Complexes (PECs). PECs can exist in many different states, depending on the architecture of the polymers and the environmental parameters of the solution. Using double hydrophilic block copolymers (DHBCs), PECs can be stabilized as dispersed aggregates in solutions. Specifically, the polymers involved in this investigation are a DHBC composed of a poly(ethylene glycol) block and a poly(methacrylic acid) block (PEO-PMAA) used as the polyanion and poly(2-(dimethylamino)ethyl methacrylate), with and without hydrophobic dodecyl substitutions, used as the polycation. In this paper, we discuss the behavior of the nanoscale dynamics with respect to their mixing ratio. We also test the impact of hydrophobic modifications on the dynamics of the aggregates. By neutron spin echo spectroscopy and neutron backscattering spectroscopy, we observed the role of electrostatic interaction as a friction induced on the polymers, where complexation leads to slower diffusion and the hydrophobic moieties affect the rigidity of the polymers