Search

Given a set of measurements, observability characterizes the distinguishability of a system's initial state, whereas constructability focuses on the final state in a trajectory. In the presence of process and/or measurement noise, the Fisher information matrices with respect to the initial and final states$\unicode{x2013}$equivalent to the stochastic observability and constructability Gramians$\unicode{x2013}$bound the performance of corresponding estimators through the Cram\'er-Rao inequality. This letter establishes a connection between stochastic observability and constructability of discrete-time linear systems and provides a more numerically stable way for calculating the stochastic observability Gramian. We define a dual system and show that the dual system's stochastic constructability is equivalent to the original system's stochastic observability, and vice versa. This duality enables the interchange of theorems and tools for observability and constructability. For example, we use this result to translate an existing recursive formula for the stochastic constructability Gramian into a formula for recursively calculating the stochastic observability Gramian for both time-varying and time-invariant systems, and we show the convergence of this sequence for the latter. Finally, we illustrate the robustness of our formula compared to existing (non-recursive) formulas through a numerical example., Comment: 10 pages, 2 figures

Unlike quasi-static robotic manipulation tasks like pick-and-place, dynamic tasks such as non-prehensile manipulation pose greater challenges, especially for vision-based control. Successful control requires the extraction of features relevant to the target task. In visual imitation learning settings, these features can be learnt by backpropagating the policy loss through the vision backbone. Yet, this approach tends to learn task-specific features with limited generalizability. Alternatively, learning world models can realize more generalizable vision backbones. Utilizing the learnt features, task-specific policies are subsequently trained. Commonly, these models are trained solely to predict the next RGB state from the current state and action taken. But only-RGB prediction might not fully-capture the task-relevant dynamics. In this work, we hypothesize that direct supervision of target dynamic states (Dynamics Mapping) can learn better dynamics-informed world models. Beside the next RGB reconstruction, the world model is also trained to directly predict position, velocity, and acceleration of environment rigid bodies. To verify our hypothesis, we designed a non-prehensile 2D environment tailored to two tasks: "Balance-Reaching" and "Bin-Dropping". When trained on the first task, dynamics mapping enhanced the task performance under different training configurations (Decoupled, Joint, End-to-End) and policy architectures (Feedforward, Recurrent). Notably, its most significant impact was for world model pretraining boosting the success rate from 21% to 85%. Although frozen dynamics-informed world models could generalize well to a task with in-domain dynamics, but poorly to a one with out-of-domain dynamics., Comment: Accepted to IEEE CASE 2024

A distribution shift between the training and test data can severely harm performance of machine learning models. Importance weighting addresses this issue by assigning different weights to data points during training. We argue that existing heuristics for determining the weights are suboptimal, as they neglect the increase of the variance of the estimated model due to the finite sample size of the training data. We interpret the optimal weights in terms of a bias-variance trade-off, and propose a bi-level optimization procedure in which the weights and model parameters are optimized simultaneously. We apply this optimization to existing importance weighting techniques for last-layer retraining of deep neural networks in the presence of sub-population shifts and show empirically that optimizing weights significantly improves generalization performance., Comment: Preprint. Currently under review

The increased brightness temperature of young rocky protoplanets during their magma ocean epoch makes them potentially amenable to atmospheric characterization to distances from the solar system far greater than thermally equilibrated terrestrial exoplanets, offering observational opportunities for unique insights into the origin of secondary atmospheres and the near surface conditions of prebiotic environments. The Large Interferometer For Exoplanets (LIFE) mission will employ a space-based mid-infrared nulling interferometer to directly measure the thermal emission of terrestrial exoplanets. Here, we seek to assess the capabilities of various instrumental design choices of the LIFE mission concept for the detection of cooling protoplanets with transient high-temperature magma ocean atmospheres, in young stellar associations in particular. Using the LIFE mission instrument simulator (LIFEsim) we assess how specific instrumental parameters and design choices, such as wavelength coverage, aperture diameter, and photon throughput, facilitate or disadvantage the detection of protoplanets. We focus on the observational sensitivities of distance to the observed planetary system, protoplanet brightness temperature using a blackbody assumption, and orbital distance of the potential protoplanets around both G- and M-dwarf stars. Our simulations suggest that LIFE will be able to detect (S/N $\geq$ 7) hot protoplanets in young stellar associations up to distances of $\approx$100 pc from the solar system for reasonable integration times (up to $\sim$hours). Detection of an Earth-sized protoplanet orbiting a solar-sized host star at 1 AU requires less than 30 minutes of integration time. M-dwarfs generally need shorter integration times. The contribution from wavelength regions $<$6 $\mu$m is important for decreasing the detection threshold and discriminating emission temperatures., Comment: 18 pages, 19 figures; accepted for publication in A&A

Expressivity and generalization are two critical aspects of graph neural networks (GNNs). While significant progress has been made in studying the expressivity of GNNs, much less is known about their generalization capabilities, particularly when dealing with the inherent complexity of graph-structured data. In this work, we address the intricate relationship between expressivity and generalization in GNNs. Theoretical studies conjecture a trade-off between the two: highly expressive models risk overfitting, while those focused on generalization may sacrifice expressivity. However, empirical evidence often contradicts this assumption, with expressive GNNs frequently demonstrating strong generalization. We explore this contradiction by introducing a novel framework that connects GNN generalization to the variance in graph structures they can capture. This leads us to propose a $k$-variance margin-based generalization bound that characterizes the structural properties of graph embeddings in terms of their upper-bounded expressive power. Our analysis does not rely on specific GNN architectures, making it broadly applicable across GNN models. We further uncover a trade-off between intra-class concentration and inter-class separation, both of which are crucial for effective generalization. Through case studies and experiments on real-world datasets, we demonstrate that our theoretical findings align with empirical results, offering a deeper understanding of how expressivity can enhance GNN generalization., Comment: 17 pages, 2 figures, 2 tables

The inhomogeneous magnetic stray field of micromagnets has been extensively used to manipulate electron spin qubits. By means of micromagnetic simulations and scanning superconducting quantum interference device microscopy, we show that the polycrystallinity of the magnet and non-uniform magnetization significantly impact the stray field and corresponding qubit properties. We find that the random orientation of the crystal axis in polycrystalline Co magnets alters the qubit frequencies by up to 0.5 GHz, compromising single qubit addressability and single gate fidelities. We map the stray field of Fe micromagnets with an applied magnetic field of up to 500 mT (mimicking conditions when operating qubits), finding field gradients above 1 mT/nm. The measured gradients and the lower magnetocrystalline anisotropy of Fe demonstrate the advantage of using Fe instead of Co for magnets in spin qubit devices. These properties of Fe also enabled us to design a 2D arrangement of nanomagnets for driving spin qubits distributed on a 2D lattice.

Graph Neural Networks (GNNs) have become an essential tool for analyzing graph-structured data, leveraging their ability to capture complex relational information. While the expressivity of GNNs, particularly their equivalence to the Weisfeiler-Leman (1-WL) isomorphism test, has been well-documented, understanding their generalization capabilities remains critical. This paper focuses on the generalization of GNNs by investigating their Vapnik-Chervonenkis (VC) dimension. We extend previous results to demonstrate that 1-dimensional GNNs with a single parameter have an infinite VC dimension for unbounded graphs. Furthermore, we show that this also holds for GNNs using analytic non-polynomial activation functions, including the 1-dimensional GNNs that were recently shown to be as expressive as the 1-WL test. These results suggest inherent limitations in the generalization ability of even the most simple GNNs, when viewed from the VC dimension perspective., Comment: 10 pages

Quantum computing has emerged as a promising tool for transforming the landscape of computing technology. Recent efforts have applied quantum techniques to classical database challenges, such as query optimization, data integration, index selection, and transaction management. In this paper, we shift focus to a critical yet underexplored area: data management for quantum computing. We are currently in the Noisy Intermediate-Scale Quantum (NISQ) era, where qubits, while promising, are fragile and still limited in scale. After differentiating quantum data from classical data, we outline current and future data management paradigms in the NISQ era and beyond. We address the data management challenges arising from the emerging demands of near-term quantum computing. Our goal is to chart a clear course for future quantum-oriented data management research, establishing it as a cornerstone for the advancement of quantum computing in the NISQ era., Comment: This vision paper is crafted to be accessible to a broad audience. We welcome any questions, feedback, or suggestions. Please feel free to reach out to us

We study DC magnetotransport in a quantum critical metal in the presence of a lattice. In the regime where the transport is hydrodynamical the interplay of the Lorentz force and the lattice gives rise to a natural anomalous contribution to the cyclotron frequency that changes it from its canonical charge-to-mass ratio. The size of this effect is universal as it is determined only by thermodynamic quantities. Remarkably the Drude weight changes in such a way that to first subleading order in the lattice strength the Hall resistivity and Hall coefficient do not change, though the Hall angle does change. We confirm our results with numerical simulations in a holographic model of a strange metal. For weak lattice strength these hydrodynamic effects are shown to be present. The numerical simulations also suggest that strong lattice effects beyond a hydrodynamic regime may provide a resolution to the experimentally observed anomalous Hall response of cuprate strange metals., Comment: 29+22 pages, 10+3 figures. v2 references corrected and added

Introduction: Thorough maintenance of the scientific record is needed to ensure the trustworthiness of its content. This can be undermined by a stealth correction, which is at least one post-publication change made to a scientific article, without providing a correction note or any other indicator that the publication was temporarily or permanently altered. In this paper we provide several examples of stealth corrections in order to demonstrate that these exist within the scientific literature. As far as we are aware, no documentation of such stealth corrections was previously reported in the scientific literature. Methods: We identified stealth corrections ourselves, or found already reported ones on the public database pubpeer.com or through social media accounts of known science sleuths. Results: In total we report 131 articles that were affected by stealth corrections and were published between 2005 and 2024. These stealth corrections were found among multiple publishers and scientific fields. Conclusion: and recommendations Stealth corrections exist in the scientific literature. This needs to end immediately as it threatens scientific integrity. We recommend the following: 1) Tracking all changes to the published record by all publishers in an open, uniform and transparent manner, preferably by online submission systems that log every change publicly, making stealth corrections impossible; 2) Clear definitions and guidelines on all types of corrections; 3) Support sustained vigilance of the scientific community to publicly register stealth corrections., Comment: Manuscript and supplemental table combined in 1 PDF

Attention is a key part of the transformer architecture. It is a sequence-to-sequence mapping that transforms each sequence element into a weighted sum of values. The weights are typically obtained as the softmax of dot products between keys and queries. Recent work has explored alternatives to softmax attention in transformers, such as ReLU and sigmoid activations. In this work, we revisit sigmoid attention and conduct an in-depth theoretical and empirical analysis. Theoretically, we prove that transformers with sigmoid attention are universal function approximators and benefit from improved regularity compared to softmax attention. Through detailed empirical analysis, we identify stabilization of large initial attention norms during the early stages of training as a crucial factor for the successful training of models with sigmoid attention, outperforming prior attempts. We also introduce FLASHSIGMOID, a hardware-aware and memory-efficient implementation of sigmoid attention yielding a 17% inference kernel speed-up over FLASHATTENTION2 on H100 GPUs. Experiments across language, vision, and speech show that properly normalized sigmoid attention matches the strong performance of softmax attention on a wide range of domains and scales, which previous attempts at sigmoid attention were unable to fully achieve. Our work unifies prior art and establishes best practices for sigmoid attention as a drop-in softmax replacement in transformers.

Sign Language Processing (SLP) provides a foundation for a more inclusive future in language technology; however, the field faces several significant challenges that must be addressed to achieve practical, real-world applications. This work addresses multi-view isolated sign recognition (MV-ISR), and highlights the essential role of 3D awareness and geometry in SLP systems. We introduce the NGT200 dataset, a novel spatio-temporal multi-view benchmark, establishing MV-ISR as distinct from single-view ISR (SV-ISR). We demonstrate the benefits of synthetic data and propose conditioning sign representations on spatial symmetries inherent in sign language. Leveraging an SE(2) equivariant model improves MV-ISR performance by 8%-22% over the baseline., Comment: Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM) at the 41 st International Conference on Machine Learning, Vienna, Austria. PMLR 251, 2024

In this work, we present an efficient approach for capturing sign language in 3D, introduce the 3D-LEX v1.0 dataset, and detail a method for semi-automatic annotation of phonetic properties. Our procedure integrates three motion capture techniques encompassing high-resolution 3D poses, 3D handshapes, and depth-aware facial features, and attains an average sampling rate of one sign every 10 seconds. This includes the time for presenting a sign example, performing and recording the sign, and archiving the capture. The 3D-LEX dataset includes 1,000 signs from American Sign Language and an additional 1,000 signs from the Sign Language of the Netherlands. We showcase the dataset utility by presenting a simple method for generating handshape annotations directly from 3D-LEX. We produce handshape labels for 1,000 signs from American Sign Language and evaluate the labels in a sign recognition task. The labels enhance gloss recognition accuracy by 5% over using no handshape annotations, and by 1% over expert annotations. Our motion capture data supports in-depth analysis of sign features and facilitates the generation of 2D projections from any viewpoint. The 3D-LEX collection has been aligned with existing sign language benchmarks and linguistic resources, to support studies in 3D-aware sign language processing.

Contamination from singly ionized iron emission is one of the greatest obstacles to determining the intensity of emission lines in the UV and optical wavelength ranges. This study presents a comprehensive analysis of the FeII emission in the bright quasar RM 102, based on the most recent version of the CLOUDY software, with the goal of simultaneously reproducing UV and optical FeII emission. We employ a constant pressure model for the emitting clouds, instead of the customary constant density assumption. The allowed parameter range is broad, with metallicity up to 50 times the solar value and turbulent velocity up to 100 km s$^{-1}$ for a subset of models. We also consider geometrical effects that could enhance the visibility of the non-illuminated faces of the clouds, as well as additional mechanical heating. Our investigation reveals that the broad line region of RM 102 is characterized by highly metallic gas. The observed FeII features provide strong evidence for an inflow pattern geometry that favours the dark sides of clouds over isotropic emission. This study confirms the presence of chemically enriched gas in the broad line region of bright quasars, represented by RM 102, which is necessary to explain the strong FeII emission and its characteristic features. Additionally, we report that CLOUDY currently still lacks certain transitions in its atomic databases which prevents it from fully reproducing some observed FeII features in quasar spectra.

Exchange bias is applied ubiquitously throughout the spintronics industry for pinning magnetic moment in an orientation robust against external magnetic field disturbances. Enabling efficient manipulation of the exchange-biased pinning direction is key to add or advance device functionality. Conventional methods for achieving this reorientation commonly require long heating times and assisting magnetic fields, thus limiting speed and versatility. Here, we demonstrate a field-free $180^\circ$ reorientation of perpendicular exchange bias in a thin film system integrating Co/Gd layers with IrMn. Single femtosecond laser pulses trigger a reversal of the pinning direction, while preserving the unidirectionality required for referencing. We formulate a theoretical microscopic framework involving a wide range of timescales that accurately reproduces the experimental outcomes. The understanding of the underlying mechanisms delivers guiding principles for designing reprogrammable spintronic devices., Comment: Manuscript revision for submission

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

We consider the incompressible and stationary Stokes equations on an infinite two-dimensional wedge with non-scaling invariant Navier-slip boundary conditions. We prove well-posedness and higher regularity of the Stokes problem in a certain class of weighted Sobolev spaces. The novelty of this work is the occurrence of two different scalings in the boundary condition, which is not treated so far for the Stokes system in unbounded wedge-type domains. These difficulties are overcome by first constructing a variational solution in a second order weighted Sobolev space and subsequently proving higher regularity up to the tip of the wedge by employing an iterative scheme. We believe that this method can be used for other problems with variational structure and multiple scales.

The Single Aperture Large Telescope for Universe Studies (SALTUS) is a far-infrared space mission concept with unprecedented spatial and spectral resolution. Saltus consists of a 14-m inflatable primary, providing 16 times the sensitivity and 4 times the angular resolution of Herschel, and two cryogenic detectors spanning a wavelength range of 34-660 microns and spectral resolving power of 300 - 1e7. Spectroscopic observations in the far-infrared offer many unique windows into the processes of star and planet formation. These include observations of low energy water transitions, the H2 mass tracer HD, many CHONS constraining molecules such as NH3 and H2S, and emission lines from the phonon modes of molecular ices. Observing these species will allow us to build a statistical sample of protoplanetary disk masses, characterize the water snowline, identify Kuiper Belt like debris rings around other stars, and trace the evolution CHONS from prestellar cores, through to protoplanetary disks and debris disks. This paper details details several key star and planet formation science goals achievable with SALTUS., Comment: 48 pages, 8 figures, submitted to SPIE JATIS

Gradient-based attacks are a primary tool to evaluate robustness of machine-learning models. However, many attacks tend to provide overly-optimistic evaluations as they use fixed loss functions, optimizers, step-size schedulers, and default hyperparameters. In this work, we tackle these limitations by proposing a parametric variation of the well-known fast minimum-norm attack algorithm, whose loss, optimizer, step-size scheduler, and hyperparameters can be dynamically adjusted. We re-evaluate 12 robust models, showing that our attack finds smaller adversarial perturbations without requiring any additional tuning. This also enables reporting adversarial robustness as a function of the perturbation budget, providing a more complete evaluation than that offered by fixed-budget attacks, while remaining efficient. We release our open-source code at https://github.com/pralab/HO-FMN.

ModSecurity is widely recognized as the standard open-source Web Application Firewall (WAF), maintained by the OWASP Foundation. It detects malicious requests by matching them against the Core Rule Set (CRS), identifying well-known attack patterns. Each rule is manually assigned a weight based on the severity of the corresponding attack, and a request is blocked if the sum of the weights of matched rules exceeds a given threshold. However, we argue that this strategy is largely ineffective against web attacks, as detection is only based on heuristics and not customized on the application to protect. In this work, we overcome this issue by proposing a machine-learning model that uses the CRS rules as input features. Through training, ModSec-Learn is able to tune the contribution of each CRS rule to predictions, thus adapting the severity level to the web applications to protect. Our experiments show that ModSec-Learn achieves a significantly better trade-off between detection and false positive rates. Finally, we analyze how sparse regularization can reduce the number of rules that are relevant at inference time, by discarding more than 30% of the CRS rules. We release our open-source code and the dataset at https://github.com/pralab/modsec-learn and https://github.com/pralab/http-traffic-dataset, respectively., Comment: arXiv admin note: text overlap with arXiv:2308.04964

In this paper, we consider the Laplace operator on the half-space with Dirichlet and Neumann boundary conditions. We prove that this operator admits a bounded $H^\infty$-calculus on Sobolev spaces with power weights measuring the distance to the boundary. These weights do not necessarily belong to the class of Muckenhoupt $A_p$ weights. We additionally study the corresponding Dirichlet and Neumann heat semigroup. It is shown that these semigroups, in contrast to the $L^p$-case, have polynomial growth. Moreover, maximal regularity results for the heat equation are derived on inhomogeneous and homogeneous weighted Sobolev spaces., Comment: 50 pages

The SALTUS Probe mission will provide a powerful far-infrared (far-IR) pointed space observatory to explore our cosmic origins and the possibility of life elsewhere. The observatory employs an innovative deployable 14-m aperture, with a sunshield that will radiatively cool the off-axis primary to <45K. This cooled primary reflector works in tandem with cryogenic coherent and incoherent instruments that span the 34 to 660 micron far-IR range at both high and moderate spectral resolutions., Comment: 49 pages, 10 figures, 3 tables, submitted to SPIE JATIS

We prove a new generalization of a theorem of Carleson, namely bounds for a generalized Carleson operator on doubling metric measure spaces. Additionally, we explicitly reduce Carleson's classical result on pointwise convergence of Fourier series to this new theorem. Both proofs are presented in great detail, suitable as a blueprint for computer verification using the current capabilities of the software package Lean. Note that even Carleson's classical result has not yet been computer-verified., Comment: 130 pages

As of 2023, over 5500 planets are known to orbit stars other than our Sun. We can measure their sizes and orbital periods, infer their masses and temperatures, and constrain their compositions. Based on these data, about 1% of extrasolar planets are potentially habitable for life as we know it, implying that of the billions of planets in our Galaxy, some may actually be inhabited, at least by microbes. However, recognizing signs of alien life forms is a major challenge for current technology, because of the wide range of conditions on extrasolar planets, and because of the wide range of forms that life may take. This chapter reviews observations of exoplanets and discusses astrobiological definitions of habitability and the likelihood of finding life beyond the Earth, both within and outside the Solar system., Comment: Book chapter, to appear in First Contact: Aliens and Humans in Contemporary Science Fiction, eds. M. van Dijk, F. Bosman, K. Glimmerveen (Springer). Aimed at audience without professional background in natural sciences

The 4D eigenvector 1 (E1) sequence has emerged as a powerful tool for organizing the observational and physical characteristics of type-1 active galactic nuclei (AGNs). In this study, we present a comprehensive analysis of the metallicity of the broad line region gas, incorporating both new data and previously published findings, to assess the presence of any trend along the sequence. We perform a multi-component analysis on the strongest UV and optical emission lines, compute $\sim 10$ diagnostic ratios, and compare them with the prediction of CLOUDY photoionization simulations, identifying a photoionization solution closest to the data. Our investigation reveals a consistent pattern along the optical plane of the E1. We observe a systematic progression in metallicity, ranging from sub-solar values to metallicity levels several times higher than solar values. These findings underscore the role of metallicity as a fundamental correlate of the 4DE1/main sequence. Extreme values of metallicity, at least several tens solar, are confirmed in low-$z$ AGNs radiating at a high Eddington ratio, although the origin of the extreme enrichment remains open to debate.

Recent advancements in multimodal large language models (MLLMs) have been noteworthy, yet, these general-domain MLLMs often fall short in their ability to comprehend and interact effectively with user interface (UI) screens. In this paper, we present Ferret-UI, a new MLLM tailored for enhanced understanding of mobile UI screens, equipped with referring, grounding, and reasoning capabilities. Given that UI screens typically exhibit a more elongated aspect ratio and contain smaller objects of interest (e.g., icons, texts) than natural images, we incorporate "any resolution" on top of Ferret to magnify details and leverage enhanced visual features. Specifically, each screen is divided into 2 sub-images based on the original aspect ratio (i.e., horizontal division for portrait screens and vertical division for landscape screens). Both sub-images are encoded separately before being sent to LLMs. We meticulously gather training samples from an extensive range of elementary UI tasks, such as icon recognition, find text, and widget listing. These samples are formatted for instruction-following with region annotations to facilitate precise referring and grounding. To augment the model's reasoning ability, we further compile a dataset for advanced tasks, including detailed description, perception/interaction conversations, and function inference. After training on the curated datasets, Ferret-UI exhibits outstanding comprehension of UI screens and the capability to execute open-ended instructions. For model evaluation, we establish a comprehensive benchmark encompassing all the aforementioned tasks. Ferret-UI excels not only beyond most open-source UI MLLMs, but also surpasses GPT-4V on all the elementary UI tasks.

We present Conformal Intent Classification and Clarification (CICC), a framework for fast and accurate intent classification for task-oriented dialogue systems. The framework turns heuristic uncertainty scores of any intent classifier into a clarification question that is guaranteed to contain the true intent at a pre-defined confidence level. By disambiguating between a small number of likely intents, the user query can be resolved quickly and accurately. Additionally, we propose to augment the framework for out-of-scope detection. In a comparative evaluation using seven intent recognition datasets we find that CICC generates small clarification questions and is capable of out-of-scope detection. CICC can help practitioners and researchers substantially in improving the user experience of dialogue agents with specific clarification questions., Comment: 9 pages,2 figures,3 tables,6 appendices,to be published in ACL's NAACL Findings 2024

In this work, we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision language connector, and various pre-training data choices, we identified several crucial design lessons. For example, we demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art (SOTA) few-shot results across multiple benchmarks, compared to other published pre-training results. Further, we show that the image encoder together with image resolution and the image token count has substantial impact, while the vision-language connector design is of comparatively negligible importance. By scaling up the presented recipe, we build MM1, a family of multimodal models up to 30B parameters, including both dense models and mixture-of-experts (MoE) variants, that are SOTA in pre-training metrics and achieve competitive performance after supervised fine-tuning on a range of established multimodal benchmarks. Thanks to large-scale pre-training, MM1 enjoys appealing properties such as enhanced in-context learning, and multi-image reasoning, enabling few-shot chain-of-thought prompting.

The Weisfeiler-Leman algorithm ($1$-WL) is a well-studied heuristic for the graph isomorphism problem. Recently, the algorithm has played a prominent role in understanding the expressive power of message-passing graph neural networks (MPNNs) and being effective as a graph kernel. Despite its success, $1$-WL faces challenges in distinguishing non-isomorphic graphs, leading to the development of more expressive MPNN and kernel architectures. However, the relationship between enhanced expressivity and improved generalization performance remains unclear. Here, we show that an architecture's expressivity offers limited insights into its generalization performance when viewed through graph isomorphism. Moreover, we focus on augmenting $1$-WL and MPNNs with subgraph information and employ classical margin theory to investigate the conditions under which an architecture's increased expressivity aligns with improved generalization performance. In addition, we show that gradient flow pushes the MPNN's weights toward the maximum margin solution. Further, we introduce variations of expressive $1$-WL-based kernel and MPNN architectures with provable generalization properties. Our empirical study confirms the validity of our theoretical findings., Comment: Accepted at ICML 2024. arXiv admin note: text overlap with arXiv:2301.11039

We study the equivariant geometry of special quadric fibrations, called Umemura quadric fibrations, as well as the maximality of their automorphism groups inside $Cr_n(\mathbb{C})$. We produce infinite families of pairwise non-conjugate maximal connected algebraic subgroups of $Cr_n(\mathbb{C})$., Comment: 24 pages, comments are welcome

Cluckers and Lipshitz have shown that real closed fields equipped with real analytic structure are o-minimal. This generalizes the well-known subanalytic structure $\mathbb{R}_{\mathrm{an}}$ on the real numbers. We extend this line of research by investigating ordered fields with real analytic structure that are not necessarily real closed. When considered in a language with a symbol for a convex valuation ring, these structures turn out to be tame as valued fields: we prove that they are $\omega$-h-minimal. Additionally, our approach gives a precise description of the induced structure on the residue field and the value group, and naturally leads to an Ax--Kochen--Ersov-theorem for fields with real analytic structure., Comment: 28 pages

The inner regions of protoplanetary disks are the locations where most of planets are thought to form and where processes that influence the global evolution of the disk, such as MHD-winds and photoevaporation, originate. Transition disks (TDs) with large inner dust cavities are the ideal targets to study the inner tens of au of disks, as the central emission can be fully disentangled from the outer disk emission. We present a homogeneous multi-wavelength analysis of the continuum emission in a sample of 11 TDs. We investigate the nature of the central emission close to the star, distinguishing between thermal dust and free-free emission. Spatially resolved measurements of continuum emission from archival ALMA data are combined with literature cm-wave observations to study the spectral indices of the inner and outer disks separately. While the emission from the outer disks is consistent with thermal dust emission, 10/11 of the spectral indices estimated for the central emission close to the star suggest that this emission is free-free emission, likely associated with an ionized jet or a disk wind. No correlation between the free-free luminosity and the accretion luminosity or the X-ray luminosity is found, arguing against the photoevaporative wind origin. A sub-linear correlation between the ionized mass loss rate and the accretion rate onto the star is observed, suggesting an origin in an ionized jet. The relative lack of mm-dust grains in the majority of inner disks in transition disks suggests that either such dust grains have drifted quickly towards the central star, grain growth is less efficient in the inner disk, or grains grow rapidly to planetesimal sizes in the inner disk. The observed correlation between the ionized mass loss rate and the accretion rate suggests the outflow is strictly connected with the stellar accretion and that accretion in these disks is driven by a jet., Comment: Accepted for publication in A&A

We introduce Physically Enhanced Gaussian Splatting Simulation System (PEGASUS) for 6DOF object pose dataset generation, a versatile dataset generator based on 3D Gaussian Splatting. Environment and object representations can be easily obtained using commodity cameras to reconstruct with Gaussian Splatting. PEGASUS allows the composition of new scenes by merging the respective underlying Gaussian Splatting point cloud of an environment with one or multiple objects. Leveraging a physics engine enables the simulation of natural object placement within a scene through interaction between meshes extracted for the objects and the environment. Consequently, an extensive amount of new scenes - static or dynamic - can be created by combining different environments and objects. By rendering scenes from various perspectives, diverse data points such as RGB images, depth maps, semantic masks, and 6DoF object poses can be extracted. Our study demonstrates that training on data generated by PEGASUS enables pose estimation networks to successfully transfer from synthetic data to real-world data. Moreover, we introduce the Ramen dataset, comprising 30 Japanese cup noodle items. This dataset includes spherical scans that captures images from both object hemisphere and the Gaussian Splatting reconstruction, making them compatible with PEGASUS., Comment: Project Page: https://meyerls.github.io/pegasus_web

The 4D Eigenvector 1 sequence has proven to be a highly effective tool for organizing observational and physical properties of type 1 active galactic nuclei (AGN). In this paper, we present multiple measurements of metallicity for the broad line region gas, from new or previously published data. We demonstrate a consistent trend along the optical plane of the E1 (also known as the quasar main sequence), defined by the line width of H$\beta$ and by a parameter measuring the prominence of singly-ionized iron emission. The trend involves an increase from sub-solar metallicity in correspondence with extreme Population B (weak FeII emission, large H$\beta$ FWHM) to metallicity several tens the solar value in correspondence with extreme Population A (very strong FeII optical emission, narrower H$\beta$ profiles). The data establish the metallicity as a correlate of the 4D E1/main sequence. If the very high metallicity gas ($Z \gtrsim 10 Z_\odot$) is expelled from the sphere of influence of the central black hole, as indicated by the widespread evidence of nuclear outflows and disk wind in the case of sources radiating at high Eddington ratio, then it is possible that the outflows from quasars played a role in chemically enriching the host galaxy., Comment: Submitted to Physics

There is a growing demand for explainable, transparent, and data-driven models within the domain of fraud detection. Decisions made by fraud detection models need to be explainable in the event of a customer dispute. Additionally, the decision-making process in the model must be transparent to win the trust of regulators and business stakeholders. At the same time, fraud detection solutions can benefit from data due to the noisy, dynamic nature of fraud and the availability of large historical data sets. Finally, fraud detection is notorious for its class imbalance: there are typically several orders of magnitude more legitimate transactions than fraudulent ones. In this paper, we present Deep Symbolic Classification (DSC), an extension of the Deep Symbolic Regression framework to classification problems. DSC casts classification as a search problem in the space of all analytic functions composed of a vocabulary of variables, constants, and operations and optimizes for an arbitrary evaluation metric directly. The search is guided by a deep neural network trained with reinforcement learning. Because the functions are mathematical expressions that are in closed-form and concise, the model is inherently explainable both at the level of a single classification decision and the model's decision process. Furthermore, the class imbalance problem is successfully addressed by optimizing for metrics that are robust to class imbalance such as the F1 score. This eliminates the need for oversampling and undersampling techniques that plague traditional approaches. Finally, the model allows to explicitly balance between the prediction accuracy and the explainability. An evaluation on the PaySim data set demonstrates competitive predictive performance with state-of-the-art models, while surpassing them in terms of explainability. This establishes DSC as a promising model for fraud detection systems., Comment: 12 pages, 3 figures, To be published in the 3rd International Workshop on Explainable AI in Finance of the 4th ACM International Conference on AI in Finance (ICAIF, https://ai-finance.org/)