arXiv subjects

Luca Rossi

Publications and source records attributed to Luca Rossi.

At least 19 recordsLinked to original sources

arXiv2026

Looking into the faintEst WIth MUSE (LEWIS): Exploring the nature of ultra-diffuse galaxies in the Hydra-I cluster. VI. A star-forming UDG in Hydra I: a rare UDG or a transition phase?

This paper presents a detailed analysis of a gas-rich star-forming ultra-diffuse galaxy (UDG) as part of the ESO Large Programme 'Looking into the faintEst WIth MUSE (LEWIS)'. Among the UDGs in the LEWIS sample, UDG 6 is the only galaxy that hosts a significant amount of ionised gas with evidence of emission lines, suggesting recent star-forming activity. The main goal of this work is to constrain the formation history of this UDG by comparing its properties with the main formation scenarios proposed for this extreme class of galaxies. We adopted integral field spectroscopy from MUSE to derive the morphology and the structural properties of the stellar and gas components of UDG 6. We applied spectral fitting and Voronoi tessellation algorithms to the MUSE data-cube to derive the kinematics and properties of the gas and stellar component. Moreover, we derived the GCs populations' properties by applying a multi-band spectrophotometric analysis. We confirmed that UDG 6 is a member of Hydra I cluster. It is characterised by a regular and elongated shape and contains a significant dust content, a metal-poor ionised gas fraction and an underlying old-to-intermediate stellar component. Evidence of local and clumpy star-forming activity has been revealed through the analysis of emission line, and an arc-like tidal feature was discovered from unsharp masking analysis. UDG 6 might originate from a 'puffed-up dwarf' whose stellar content has been stretched out to larger radii, passively evolving into a more diffuse galaxy. Being located in a dynamically active region of the cluster, characterised by tidal features and stripping phenomena, we suggest that the environmental processes have played a role in shaping the properties of UDG 6. A tidal interaction with a nearby galaxy might have triggered recent star-formation activity, without dramatically altering the coherent gas rotation in UDG 6.

astro-ph.GA↗

arXiv2026

Freidlin-G\"artner formula and asymptotic profile in reaction-diffusion equations

We address the question of the large-time behavior of solutions to reaction-diffusion equations in periodic media. We start with the description of the asymptotic shape of the invasion set, which is characterized by the Freidlin-G\"artner formula. We outline a proof of the formula that holds true for general types of reaction terms. We then present some recent results, obtained in collaboration with H. Guo and F. Hamel, for (weakly) bistable equations. They include a regular version of the Freidlin-G\"artner formula and the convergence in profile towards pulsating traveling fronts for solutions with either bounded or unbounded initial support.

math.AP↗

arXiv2026

SP-GCRL: Influence Maximization on Incomplete Social Graphs

Influence maximization (IM) in real platforms is challenged by incomplete, noisy social graphs and non-stationary diffusion dynamics. We propose SP-GCRL, a social-propagation-aware graph contrastive reinforcement learning framework that learns end-to-end seed selection under partial observability.We first introduce a social-propagation-aware nonlinear diffusion function to model reinforcement/diminishing effects and probability drift under repeated exposure; we then construct dual structural views and perform contrastive learning to obtain node representations robust to missing edges and weak ties, while replacing expensive strategy metrics with a GAT-based regression surrogate to improve efficiency and scalability; finally, we use DDQN to learn an end-to-end seed selection policy on top of these representations. Experiments on multiple real-world networks show that SP-GCRL achieves significant gains over heuristic and learning-based baselines across budgets and topologies, while maintaining strong large-scale scalability.

cs.SI↗

arXiv2026

An SO(3)-equivariant reciprocal-space neural potential for long-range interactions

Long-range electrostatic and polarization interactions play a central role in molecular and condensed-phase systems, yet remain fundamentally incompatible with locality-based machine-learning interatomic potentials. Although modern SO(3)-equivariant neural potentials achieve high accuracy for short-range chemistry, they cannot represent the anisotropic, slowly decaying multipolar correlations governing realistic materials, while existing long-range extensions either break SO(3) equivariance or fail to maintain energy-force consistency. Here we introduce EquiEwald, a unified neural interatomic potential that embeds an Ewald-inspired reciprocal-space formulation within an irreducible SO(3)-equivariant framework. By performing equivariant message passing in reciprocal space through learned equivariant k-space filters and an equivariant inverse transform, EquiEwald captures anisotropic, tensorial long-range correlations without sacrificing physical consistency. Across periodic and aperiodic benchmarks, EquiEwald captures long-range electrostatic behavior consistent with ab initio reference data and consistently improves energy and force accuracy, data efficiency, and long-range extrapolation. These results establish EquiEwald as a physically principled paradigm for long-range-capable machine-learning interatomic potentials.

physics.chem-ph↗

arXiv2026

Is Seeing Believing? Evaluating Human Sensitivity to Synthetic Video

Advances in machine learning have enabled the creation of realistic synthetic videos known as deepfakes. As deepfakes proliferate, concerns about rapid spread of disinformation and manipulation of public perception are mounting. Despite the alarming implications, our understanding of how individuals perceive synthetic media remains limited, obstructing the development of effective mitigation strategies. This paper aims to narrow this gap by investigating human responses to visual and auditory distortions of videos and deepfake-generated visuals and narration. In two between-subjects experiments, we study whether audio-visual distortions affect cognitive processing, such as subjective credibility assessment and objective learning outcomes. A third study reveals that artifacts from deepfakes influence credibility. The three studies show that video distortions and deepfake artifacts can reduce credibility. Our research contributes to the ongoing exploration of the cognitive processes involved in the evaluation and perception of synthetic videos, and underscores the need for further theory development concerning deepfake exposure.

cs.HC↗

arXiv2026

On the criticality and the principal eigenvalue of almost periodic elliptic operators

We review the notion and the properties of the generalised \pe\ for elliptic operators in unbounded domains, and we relate it with the criticality theory. We focus on operators with almost periodic coefficients. We present a Liouville-type result in dimension $N\leq2$. Next, we show with a counter-example that criticality is not equivalent to the existence of an almost periodic principal eigenvalue, even for self-adjoint operators. Finally, we exhibit an almost periodic operator which is subcritical but which admits a critical limit operator. This is a manifestation of the instability character of the criticality property in the almost periodic setting.

math.AP↗

arXiv2026

LoRAP: Low-Rank Aggregation Prompting for Quantized Graph Neural Networks Training

Graph Neural Networks (GNNs) are neural networks that aim to process graph data, capturing the relationships and interactions between nodes using the message-passing mechanism. GNN quantization has emerged as a promising approach for reducing model size and accelerating inference in resource-constrained environments. Compared to quantization in LLMs, quantizing graph features is more emphasized in GNNs. Inspired by the above, we propose to leverage prompt learning, which manipulates the input data, to improve the performance of quantization-aware training (QAT) for GNNs. To mitigate the issue that prompting the node features alone can only make part of the quantized aggregation result optimal, we introduce Low-Rank Aggregation Prompting (LoRAP), which injects lightweight, input-dependent prompts into each aggregated feature to optimize the results of quantized aggregations. Extensive evaluations on 4 leading QAT frameworks over 9 graph datasets demonstrate that LoRAP consistently enhances the performance of low-bit quantized GNNs while introducing a minimal computational overhead.

cs.LG↗

arXiv2025

Looking into the faintEst WIth MUSE (LEWIS): on the nature of ultra-diffuse galaxies in the Hydra I cluster. V. Integrated stellar population properties

Context: This paper presents new results from the ESO Large Programme Looking into the faintEst WIth MUSE (LEWIS). The sample consists of low-surface brightness (LSB) and ultra-diffuse galaxies (UDGs) located inside 0.4 R$_{vir}$ of the Hydra I cluster. Integral field spectroscopy is acquired for 24 galaxies with the MUSE spectrograph mounted on the Very Large Telescope (VLT). Aims: Our main objective is to analyse possible correlations between the environment and the integrated stellar population properties. Methods: For each galaxy, we extract the 1D stacked spectrum in an aperture of one effective radius and adopt previously published stellar kinematics to derive age, metallicity and [Mg/Fe] through a full spectral fitting technique. Results: We find that the analysed LEWIS sample has a mean metallicity of [M/H] = -0.9 dex, a mean age of 10 Gyr, comparable to results of UDGs in other clusters. According to the projected phase-space, galaxies can be classified into two groups: early infallers, with slightly higher metallicities ([M/H]$_{early} = -0.8 \pm 0.1$ dex), and late infallers, with slightly lower values ([M/H]$_{late} = -1.0 \pm 0.1$ dex). Two types of galaxies are found in the early-infall region: half of them have metallicities consistent with the dwarf mass-metallicity relation, while the other half show higher values. Moreover, they display different timescales for stellar mass assembly. Conclusions: We identified different classes of UDGs in the Hydra I cluster, which suggest different formation mechanisms taking place. Almost all of the UDGs and LSBs in the cluster are consistent with the puffed-up dwarf formation scenario, having dwarf-like metallicities and being consistent with the dwarf mass-metallicity relation. In the innermost regions of the cluster, where metal-richer UDGs lie, tidal or environmental effects might have influenced their formation and evolution.

astro-ph.GA↗

arXiv2025

Looking into the faintEst WIth MUSE (LEWIS): Exploring the nature of ultra-diffuse galaxies in the Hydra-I cluster IV. A study of the Globular Cluster population in four UDGs

As old stellar systems, globular clusters (GCs) are key fossil tracers of galaxy formation and interaction histories. This paper is part of the LEWIS project, an integral-field spectroscopic survey of ultra-diffuse galaxies (UDGs) in the Hydra I cluster. We use MUSE spectroscopy and new VIRCAM $H$-band imaging data to study the GC populations and dark matter content in four dwarf galaxies. We retrieved line-of-sight velocities for all sources in the observed MUSE fields. Since the spectroscopic measurements are limited to relatively bright sources, we developed a multi-band photometric procedure to identify additional GC candidates too faint for spectroscopic confirmation. GC candidates were selected using a combination of photometric properties and morphometric criteria. Additionally, the $H$-band observations were used to constrain the stellar masses of the studied galaxies. Based on the spectroscopic classification, we confirm one GC in UDG3, two in UDG7, and four in UDG11, while UDG9 has no spectroscopically confirmed bright GCs. We identify four intra-cluster GCs in the vicinity of UDG3 and UDG11, and one ultra-compact dwarf with a radial velocity only $\Delta v = -85 \pm 10\mathrm{km\ s^{-1}}$ relative to UDG7, suggesting it may be bound to it. Considering completeness corrections and accounting for possible contamination, from photometry we estimate that the number of GCs ranges between 0 and $\sim40$ for the investigated UDGs. Their specific frequencies suggest that three out of four UDGs are either GC-rich, similar to those in the Coma cluster, or belong to an intermediate population as seen in the Perseus cluster. Dark matter content estimates, inferred from GC counts and stellar mass, indicate that these galaxies are dark-matter dominated, with dynamical-to-stellar mass ratios of $M_{\mathrm{dyn}} / M_\star \sim 10-1000$.

astro-ph.GA↗

arXiv2025

Reaction-diffusion equations in periodic media: convergence to pulsating fronts

This paper is concerned with reaction-diffusion-advection equations in spatially periodic media. Under an assumption of weak stability of the constant states 0 and 1, and of existence of pulsating traveling fronts connecting them, we show that fronts' profiles appear, along sequences of times and points, in the large-time dynamics of the solutions of the Cauchy problem, whether their initial supports are bounded or unbounded. The types of equations that fit into our assumptions are the combustion and the bistable ones. We also show a generalized Freidlin-G{\"a}rtner formula and other geometrical properties of the asymptotic invasion shapes, or spreading sets, of invading solutions, and we relate these sets to the upper level sets of the solutions.

math.AP↗

arXiv2025

RealHarm: A Collection of Real-World Language Model Application Failures

Language model deployments in consumer-facing applications introduce numerous risks. While existing research on harms and hazards of such applications follows top-down approaches derived from regulatory frameworks and theoretical analyses, empirical evidence of real-world failure modes remains underexplored. In this work, we introduce RealHarm, a dataset of annotated problematic interactions with AI agents built from a systematic review of publicly reported incidents. Analyzing harms, causes, and hazards specifically from the deployer's perspective, we find that reputational damage constitutes the predominant organizational harm, while misinformation emerges as the most common hazard category. We empirically evaluate state-of-the-art guardrails and content moderation systems to probe whether such systems would have prevented the incidents, revealing a significant gap in the protection of AI applications.

cs.CY↗

arXiv2025

BioX-CPath: Biologically-driven Explainable Diagnostics for Multistain IHC Computational Pathology

The development of biologically interpretable and explainable models remains a key challenge in computational pathology, particularly for multistain immunohistochemistry (IHC) analysis. We present BioX-CPath, an explainable graph neural network architecture for whole slide image (WSI) classification that leverages both spatial and semantic features across multiple stains. At its core, BioX-CPath introduces a novel Stain-Aware Attention Pooling (SAAP) module that generates biologically meaningful, stain-aware patient embeddings. Our approach achieves state-of-the-art performance on both Rheumatoid Arthritis and Sjogren's Disease multistain datasets. Beyond performance metrics, BioX-CPath provides interpretable insights through stain attention scores, entropy measures, and stain interaction scores, that permit measuring model alignment with known pathological mechanisms. This biological grounding, combined with strong classification performance, makes BioX-CPath particularly suitable for clinical applications where interpretability is key. Source code and documentation can be found at: https://github.com/AmayaGS/BioX-CPath.

cs.CV↗

arXiv2025

PHGNN: A Novel Prompted Hypergraph Neural Network to Diagnose Alzheimer's Disease

The accurate diagnosis of Alzheimer's disease (AD) and prognosis of mild cognitive impairment (MCI) conversion are crucial for early intervention. However, existing multimodal methods face several challenges, from the heterogeneity of input data, to underexplored modality interactions, missing data due to patient dropouts, and limited data caused by the time-consuming and costly data collection process. In this paper, we propose a novel Prompted Hypergraph Neural Network (PHGNN) framework that addresses these limitations by integrating hypergraph based learning with prompt learning. Hypergraphs capture higher-order relationships between different modalities, while our prompt learning approach for hypergraphs, adapted from NLP, enables efficient training with limited data. Our model is validated through extensive experiments on the ADNI dataset, outperforming SOTA methods in both AD diagnosis and the prediction of MCI conversion.

cs.LG↗

arXiv2025

SuperCap: Multi-resolution Superpixel-based Image Captioning

It has been a longstanding goal within image captioning to move beyond a dependence on object detection. We investigate using superpixels coupled with Vision Language Models (VLMs) to bridge the gap between detector-based captioning architectures and those that solely pretrain on large datasets. Our novel superpixel approach ensures that the model receives object-like features whilst the use of VLMs provides our model with open set object understanding. Furthermore, we extend our architecture to make use of multi-resolution inputs, allowing our model to view images in different levels of detail, and use an attention mechanism to determine which parts are most relevant to the caption. We demonstrate our model's performance with multiple VLMs and through a range of ablations detailing the impact of different architectural choices. Our full model achieves a competitive CIDEr score of $136.9$ on the COCO Karpathy split.

cs.CV↗

arXiv2025

Stability of propagating terraces in spatially periodic multistable equations in $\mathbb{R}^N$

In this paper, we study the large time behaviour of solutions of multistable reaction-diffusion equations in $\mathbb{R}^N$, with a spatially periodic heterogeneity. By multistable, we mean that the problem admits a finite -- but arbitrarily large -- number of stable, periodic steady states. In contrast with the more classical monostable and bistable frameworks, which exhibit the emergence of a single travelling front in the long run, in the present case the large time dynamics is governed by a family of stacked travelling fronts, involving intermediate steady states, called propagating terrace. Their existence in the multidimensional case has been established in our previous work [13]. The first result of the present paper is their uniqueness. Next, we show that the speeds of the propagating terraces in different directions dictate the spreading speeds of solutions of the Cauchy problem, for both planar-like and compactly supported initial data. The latter case turns out to be much more intricate than the former, due to the fact that the propagating terraces in distinct directions may involve different sets of intermediate steady states. Another source of difficulty is that the Wulff shape of the speeds of travelling fronts can be non-smooth, as we show in the bistable case using a result of [4].

math.AP↗

arXiv2025

TikTok StitchGraph: Characterizing communication patterns on TikTok through a collection of interaction networks

We present TikTok StitchGraph: a collection of 36 graphs based on TikTok stitches. With its rapid growth and widespread popularity, TikTok presents a compelling platform for study, yet given its video-first nature the network structure of the conversations that it hosts remains largely unexplored. Leveraging its recently released APIs, in combination with web scraping, we construct graphs detailing stitch relations from both a video- and user-centric perspective. Specifically, we focus on user multi-digraphs, with vertices representing users and edges representing directed stitch relations. From the user graphs, we characterize common communication patterns of the stitch using frequent subgraph mining, finding a preference for stars and star-like structures, an aversion towards cyclic structures, and directional disposition favoring in- and out-stars over mixed-direction structures. These structures are augmented with sentiment labels in the form of edge attributes. We then use these subgraphs for graph-level embeddings together with Graph2Vec, we show no clear distinction between topologies for different hashtag topic categories. Lastly, we compare our StitchGraphs to Twitter reply networks and show that a remakable similarity between the conversation networks on the two platforms.

cs.SI↗

arXiv2025

Generalized principal eigenvalues for parabolic operators in bounded domains

We introduce here new generalized principal eigenvalues for linear parabolic operators with heterogeneous coefficients in space and time. We consider a bounded spatial domain and an unbounded time interval $I$ : $I=\mathbb{R},\ \mathbb{R}^+$ or $\mathbb{R}^-$, and operators with coefficients having a fairly general dependence on space and time. The notions we introduce rely on the parabolic maximum principle and extend some earlier definitions introduced for elliptic operators [BNV]. We first show that these eigenvalues hold the key to understanding the large time behavior and entire solutions of heterogeneous Fisher-KPP type equations. We then describe the relation of these quantities with principal Floquet bundles for parabolic operators which provides further characterizations of the principal eigenvalues. These allow us to derive monotonicity properties and comparisons between generalized principal eigenvalues, as well as perturbation results and further properties involving limit operators. We show that the sign of these eigenvalues encodes different versions of the maximum principle for parabolic operators. Lastly, we explicitly compute the generalized principal eigenvalues for several classes of operators such as spatial-independent, periodic, almost periodic, uniquely ergodic or random stationary ergodic coefficients.

math.AP↗

arXiv2025

Looking into the faintEst WIth MUSE (LEWIS): Exploring the nature of ultra-diffuse galaxies in the Hydra-I cluster II. Stellar kinematics and dynamical masses

Context: This paper focuses on a class of galaxies characterised by an extremely low surface brightness: the ultra-diffuse galaxies (UDGs). We used new integral-field spectroscopic data from the ESO Large Programme Looking into the faintEst WIth MUSE (LEWIS) project. Aims: Our main goals are addressing the formation channels and investigating possible correlations of their observational properties. In particular, we derive their stellar kinematics and dynamical properties. Methods: We extract the 1D stacked spectrum inside the effective radius to obtain an unbiased measure of $\sigma_{\rm eff}$. To derive the spatially-resolved stellar kinematics, we first apply the Voronoi tessellation algorithm to bin the spaxels in the datacube and then follow the same prescription adopted for the 1D case. In addition, we extract the velocity profiles along the galaxy's major and minor axes. Results: We find that 7 out of 18 UDGs in LEWIS show a mild rotation, 5 do not have evidence of any rotation, and the remaining 6 UDGs are unconstrained cases. This is the first large census of velocity profiles for UDGs. On average, UDGs in LEWIS are characterised by low values of $\sigma_{\rm eff}$, comparable with available values from the literature. In the Faber-Jackson relation plane, we found a group of UDGs consistent with the relation within the errorbars, whereas outliers are objects with non-negligible rotation components. UDGs and LSBs in LEWIS have larger dark matter content than dwarf galaxies with similar total luminosity. We do not find clear correlations between the derived properties and the local environment. Conclusions: Based on the stellar kinematics, two classes of UDGs are found in the Hydra I cluster: the rotating and non-rotating systems. This result, combined with other structural properties, can help to discriminate between the several formation scenarios proposed for UDGs.

astro-ph.GA↗