We present a benchmark designed to evaluate the predictive capabilities of universal machine learning interatomic potentials across systems of varying dimensionality. Specifically, our benchmark tests zero- (molecules, atomic clusters, etc.), one- (nanowires, nanoribbons, nanotubes, etc.), two- (atomic layers and slabs) and three-dimensional (bulk materials) compounds. The benchmark reveals that while all tested models demonstrate excellent performance for three-dimensional systems, accuracy degrades progressively for lower-dimensional structures. The best performing models for geometry optimization are orbital version 2, equiformerV2, and the equivariant Smooth Energy Network, with the equivariant Smooth Energy Network also providing the most accurate energies. Our results indicate that the best models yield, on average, errors in the atomic positions in the range of 0.01-0.02 angstrom and errors in the energy below 10~meV/atom across all dimensionalities. These results demonstrate that state-of-the-art universal machine learning interatomic potentials have reached sufficient accuracy to serve as direct replacements for density functional theory calculations, at a small fraction of the computational cost, in simulations spanning the full range from isolated atoms to bulk solids. More significantly, the best performing models already enable efficient simulations of complex systems containing subsystems of mixed dimensionality, opening new possibilities for modeling realistic materials and interfaces.

With recent text-to-image models, anyone can generate deceptively realistic images with arbitrary contents, fueling the growing threat of visual disinformation. A key enabler for generating high-resolution images with low computational cost has been the development of latent diffusion models (LDMs). In contrast to conventional diffusion models, LDMs perform the denoising process in the low-dimensional latent space of a pre-trained autoencoder (AE) instead of the high-dimensional image space. Despite their relevance, the forensic analysis of LDMs is still in its infancy. In this work we propose AEROBLADE, a novel detection method which exploits an inherent component of LDMs: the AE used to transform images between image and latent space. We find that generated images can be more accurately reconstructed by the AE than real images, allowing for a simple detection approach based on the reconstruction error. Most importantly, our method is easy to implement and does not require any training, yet nearly matches the performance of detectors that rely on extensive training. We empirically demonstrate that AEROBLADE is effective against state-of-the-art LDMs, including Stable Diffusion and Midjourney. Beyond detection, our approach allows for the qualitative analysis of images, which can be leveraged for identifying inpainted regions. We release our code and data at this https URL .

Neural collapse, i.e., the emergence of highly symmetric, class-wise clustered representations, is frequently observed in deep networks and is often assumed to reflect or enable generalization. In parallel, flatness of the loss landscape has been theoretically and empirically linked to generalization. Yet, the causal role of either phenomenon remains unclear: Are they prerequisites for generalization, or merely by-products of training dynamics? We disentangle these questions using grokking, a training regime in which memorization precedes generalization, allowing us to temporally separate generalization from training dynamics and we find that while both neural collapse and relative flatness emerge near the onset of generalization, only flatness consistently predicts it. Models encouraged to collapse or prevented from collapsing generalize equally well, whereas models regularized away from flat solutions exhibit delayed generalization, resembling grokking, even in architectures and datasets where it does not typically occur. Furthermore, we show theoretically that neural collapse leads to relative flatness under classical assumptions, explaining their empirical co-occurrence. Our results support the view that relative flatness is a potentially necessary and more fundamental property for generalization, and demonstrate how grokking can serve as a powerful probe for isolating its geometric underpinnings.

Cognitive radar has emerged as a key paradigm for next-generation sensing, enabling adaptive, intelligent operation in dynamic and complex environments. Yet, conventional cognitive multiple-input multiple-output (MIMO) radars offer strong detection performance but suffer from high hardware complexity and power demands. To overcome these limitations, we develop a reinforcement learning (RL)-based framework that leverages a transmissive reconfigurable intelligent surface (TRIS) for adaptive beamforming. A state-action-reward-state-action (SARSA) agent tunes TRIS phase shifts to improve multi-target detection in low signal-to-noise ratio (SNR) conditions while operating with far fewer radio frequency (RF) chains. Simulations confirm that the proposed TRIS-RL radar matches or, for large number of elements, even surpasses MIMO performance with reduced cost and energy requirements.

Integrating watermarking into the generation process of latent diffusion models (LDMs) simplifies detection and attribution of generated content. Semantic watermarks, such as Tree-Rings and Gaussian Shading, represent a novel class of watermarking techniques that are easy to implement and highly robust against various perturbations. However, our work demonstrates a fundamental security vulnerability of semantic watermarks. We show that attackers can leverage unrelated models, even with different latent spaces and architectures (UNet vs DiT), to perform powerful and realistic forgery attacks. Specifically, we design two watermark forgery attacks. The first imprints a targeted watermark into real images by manipulating the latent representation of an arbitrary image in an unrelated LDM to get closer to the latent representation of a watermarked image. We also show that this technique can be used for watermark removal. The second attack generates new images with the target watermark by inverting a watermarked image and re-generating it with an arbitrary prompt. Both attacks just need a single reference image with the target watermark. Overall, our findings question the applicability of semantic watermarks by revealing that attackers can easily forge or remove these watermarks under realistic conditions.

Gliomas are the most common primary brain malignancies, with different degrees of aggressiveness, variable prognosis and various heterogeneous histologic sub-regions, i.e., peritumoral edematous/invaded tissue, necrotic core, active and non-enhancing core. This intrinsic heterogeneity is also portrayed in their radio-phenotype, as their sub-regions are depicted by varying intensity profiles disseminated across multi-parametric magnetic resonance imaging (mpMRI) scans, reflecting varying biological properties. Their heterogeneous shape, extent, and location are some of the factors that make these tumors difficult to resect, and in some cases inoperable. The amount of resected tumor is a factor also considered in longitudinal scans, when evaluating the apparent tumor for potential diagnosis of progression. Furthermore, there is mounting evidence that accurate segmentation of the various tumor sub-regions can offer the basis for quantitative image analysis towards prediction of patient overall survival. This study assesses the state-of-the-art machine learning (ML) methods used for brain tumor image analysis in mpMRI scans, during the last seven instances of the International Brain Tumor Segmentation (BraTS) challenge, i.e., 2012-2018. Specifically, we focus on i) evaluating segmentations of the various glioma sub-regions in pre-operative mpMRI scans, ii) assessing potential tumor progression by virtue of longitudinal growth of tumor sub-regions, beyond use of the RECIST/RANO criteria, and iii) predicting the overall survival from pre-operative mpMRI scans of patients that underwent gross total resection. Finally, we investigate the challenge of identifying the best ML algorithms for each of these tasks, considering that apart from being diverse on each instance of the challenge, the multi-institutional mpMRI BraTS dataset has also been a continuously evolving/growing dataset.

In this work, we explore the use of compact latent representations with learned time dynamics ('World Models') to simulate physical systems. Drawing on concepts from control theory, we propose a theoretical framework that explains why projecting time slices into a low-dimensional space and then concatenating to form a history ('Tokenization') is so effective at learning physics datasets, and characterise when exactly the underlying dynamics admit a reconstruction mapping from the history of previous tokenized frames to the next. To validate these claims, we develop a sequence of models with increasing complexity, starting with least-squares regression and progressing through simple linear layers, shallow adversarial learners, and ultimately full-scale generative adversarial networks (GANs). We evaluate these models on a variety of datasets, including modified forms of the heat and wave equations, the chaotic regime 2D Kuramoto-Sivashinsky equation, and a challenging computational fluid dynamics (CFD) dataset of a 2D Kármán vortex street around a fixed cylinder, where our model is successfully able to recreate the flow.

Motivated by the growing interest in integrated sensing and communication for 6th generation (6G) networks, this paper presents a cognitive Multiple-Input Multiple-Output (MIMO) radar system enhanced by reinforcement learning (RL) for robust multitarget detection in dynamic environments. The system employs a planar array configuration and adapts its transmitted waveforms and beamforming patterns to optimize detection performance in the presence of unknown two-dimensional (2D) disturbances. A robust Wald-type detector is integrated with a SARSA-based RL algorithm, enabling the radar to learn and adapt to complex clutter environments modeled by a 2D autoregressive process. Simulation results demonstrate significant improvements in detection probability compared to omnidirectional methods, particularly for low Signal-to-Noise Ratio (SNR) targets masked by clutter.

System prompts in Large Language Models (LLMs) are predefined directives that guide model behaviour, taking precedence over user inputs in text processing and generation. LLM deployers increasingly use them to ensure consistent responses across contexts. While model providers set a foundation of system prompts, deployers and third-party developers can append additional prompts without visibility into others' additions, while this layered implementation remains entirely hidden from end-users. As system prompts become more complex, they can directly or indirectly introduce unaccounted for side effects. This lack of transparency raises fundamental questions about how the position of information in different directives shapes model outputs. As such, this work examines how the placement of information affects model behaviour. To this end, we compare how models process demographic information in system versus user prompts across six commercially available LLMs and 50 demographic groups. Our analysis reveals significant biases, manifesting in differences in user representation and decision-making scenarios. Since these variations stem from inaccessible and opaque system-level configurations, they risk representational, allocative and potential other biases and downstream harms beyond the user's ability to detect or correct. Our findings draw attention to these critical issues, which have the potential to perpetuate harms if left unexamined. Further, we argue that system prompt analysis must be incorporated into AI auditing processes, particularly as customisable system prompts become increasingly prevalent in commercial AI deployments.

Tensors are fundamental in mathematics, computer science, and physics. Their study through algebraic geometry and representation theory has proved very fruitful in the context of algebraic complexity theory and quantum information. In particular, moment polytopes have been understood to play a key role. In quantum information, moment polytopes (also known as entanglement polytopes) provide a framework for the single-particle quantum marginal problem and offer a geometric characterization of entanglement. In algebraic complexity, they underpin quantum functionals that capture asymptotic tensor relations. More recently, moment polytopes have also become foundational to the emerging field of scaling algorithms in computer science and optimization. Despite their fundamental role and interest from many angles, much is still unknown about these polytopes, and in particular for tensors beyond $\mathbb{C}^2\otimes\mathbb{C}^2\otimes\mathbb{C}^2$ and $\mathbb{C}^2\otimes\mathbb{C}^2\otimes\mathbb{C}^2\otimes\mathbb{C}^2$ only sporadically have they been computed. We give a new algorithm for computing moment polytopes of tensors (and in fact moment polytopes for the general class of reductive algebraic groups) based on a mathematical description by Franz (J. Lie Theory 2002). This algorithm enables us to compute moment polytopes of tensors of dimension an order of magnitude larger than previous methods, allowing us to compute with certainty, for the first time, all moment polytopes of tensors in $\mathbb{C}^3\otimes\mathbb{C}^3\otimes\mathbb{C}^3$, and with high probability those in $\mathbb{C}^4\otimes\mathbb{C}^4\otimes\mathbb{C}^4$ (which includes the $2\times 2$ matrix multiplication tensor). We discuss how these explicit moment polytopes have led to several new theoretical directions and results.

Ultra-diffuse Galaxies (UDGs) are a subset of Low Surface Brightness Galaxies (LSBGs), showing mean effective surface brightness fainter than $24\ \rm mag\ \rm arcsec^{-2}$ and a diffuse morphology, with effective radii larger than 1.5 kpc. Due to their elusiveness, traditional methods are challenging to be used over large sky areas. Here we present a catalog of ultra-diffuse galaxy (UDG) candidates identified in the full 1350 deg$^2$ area of the Kilo-Degree Survey (KiDS) using deep learning. In particular, we use a previously developed network for the detection of low surface brightness systems in the Sloan Digital Sky Survey \citep[LSBGnet,][]{su2024lsbgnet} and optimised for UDG detection. We train this new UDG detection network for KiDS (UDGnet-K), with an iterative approach, starting from a small-scale training sample. After training and validation, the UGDnet-K has been able to identify $\sim3300$ UDG candidates, among which, after visual inspection, we have selected 545 high-quality ones. The catalog contains independent re-discovery of previously confirmed UDGs in local groups and clusters (e.g NGC 5846 and Fornax), and new discovered candidates in about 15 local systems, for a total of 67 {\it bona fide} associations. Besides the value of the catalog {\it per se} for future studies of UDG properties, this work shows the effectiveness of an iterative approach to training deep learning tools in presence of poor training samples, due to the paucity of confirmed UDG examples, which we expect to replicate for upcoming all-sky surveys like Rubin Observatory, Euclid and the China Space Station Telescope.

Gaussian states are widely regarded as one of the most relevant classes of continuous-variable (CV) quantum states, as they naturally arise in physical systems and play a key role in quantum technologies. This motivates a fundamental question: given copies of an unknown CV state, how can we efficiently test whether it is Gaussian? We address this problem from the perspective of representation theory and quantum learning theory, characterizing the sample complexity of Gaussianity testing as a function of the number of modes. For pure states, we prove that just a constant number of copies is sufficient to decide whether the state is exactly Gaussian. We then extend this to the tolerant setting, showing that a polynomial number of copies suffices to distinguish states that are close to Gaussian from those that are far. In contrast, we establish that testing Gaussianity of general mixed states necessarily requires exponentially many copies, thereby identifying a fundamental limitation in testing CV systems. Our approach relies on rotation-invariant symmetries of Gaussian states together with the recently introduced toolbox of CV trace-distance bounds.

Solana has quickly emerged as a popular platform for building decentralized applications (DApps), such as marketplaces for non-fungible tokens (NFTs). A key reason for its success are Solana's low transaction fees and high performance, which is achieved in part due to its stateless programming model. Although the literature features extensive tooling support for smart contract security, current solutions are largely tailored for the Ethereum Virtual Machine. Unfortunately, the very stateless nature of Solana's execution environment introduces novel attack patterns specific to Solana requiring a rethinking for building vulnerability analysis methods. In this paper, we address this gap and propose FuzzDelSol, the first binary-only coverage-guided fuzzing architecture for Solana smart contracts. FuzzDelSol faithfully models runtime specifics such as smart contract interactions. Moreover, since source code is not available for the large majority of Solana contracts, FuzzDelSol operates on the contract's binary code. Hence, due to the lack of semantic information, we carefully extracted low-level program and state information to develop a diverse set of bug oracles covering all major bug classes in Solana. Our extensive evaluation on 6049 smart contracts shows that FuzzDelSol's bug oracles find bugs with a high precision and recall. To the best of our knowledge, this is the largest evaluation of the security landscape on the Solana mainnet.