Search

Motivated by studies investigating causal effects in survival analysis, we propose a transformation model to quantify the impact of a binary treatment on a time-to-event outcome. The approach is based on a flexible linear transformation structural model that links a monotone function of the time-to-event with the propensity for treatment through a bivariate Gaussian distribution. The model equations are specified as functions of additive predictors, allowing the impacts of observed confounders to be accounted for flexibly. Furthermore, the effect of the instrumental variable may be regularized through a ridge penalty, while interactions between the treatment and modifier variables can be incorporated into the model to acknowledge potential variations in treatment effects across different subgroups. The baseline survival function is estimated in a flexible manner using monotonic P-splines, while unobserved confounding is captured through the dependence parameter of the bivariate Gaussian. Parameter estimation is achieved via a computationally efficient and stable penalized maximum likelihood estimation approach and intervals constructed using the related inferential results. We revisit a dataset from the Illinois Reemployment Bonus Experiment to estimate the causal effect of a cash bonus on unemployment duration, unveiling new insights. The modeling framework is incorporated into the R package GJRM, enabling researchers and practitioners to fit the proposed causal survival model and obtain easy-to-interpret numerical and visual summaries.

We propose a physics-informed data-driven framework for urban wind estimation. This framework validates and incorporates the Reynolds number independence for turbulent flows, thus allowing the extrapolation for wind conditions far beyond the training data. Another key enabler is a machine-learned non-dimensionalized manifold from snapshot data. The velocity field is modeled using a double encoder-decoder approach. The first encoder normalizes data using the oncoming wind speed, while the second encoder projects this normalized data onto the isometric feature mapping manifold. The decoders reverse this process, with $k$-nearest neighbor performing the first decoding and the second undoing the normalization. The manifold is coarse-grained by clustering to reduce the computational load for de- and encoding. The sensor-based flow estimation is based on the estimate of the oncoming wind speed and a mapping from sensor signal to the manifold latent variables. The proposed machine-learned flow estimation framework is exemplified for the flow above an Unmanned Aerial Vehicle (UAV) vertiport. The wind estimation is shown to generalize well for rare wind conditions, not included in the original database.

Measurements of the redshift drift -- the real time variation of the redshift of distance sources -- are expected in the next couple of decades using next generation facilities such as the ANDES spectrograph at the ELT and the SKAO survey. The unprecedented precision of such observations will demand precise theoretical and numerical modeling of the effect in the standard $\Lambda$CDM cosmology. In this work, we use the Gadget4 $N$-body code to simulate the redshift drift and its fluctuations in $\Lambda$CDM cosmologies, deriving the corresponding power spectra from a simulation with $1024^3$ particles in a $1\textrm{Gpc}\,h^{-1}$ box. Our results provide an estimate for the distribution and amplitude of the fluctuations and the spectra, which match previous work in the literature using Einstein-Boltzmann solvers to within an order of magnitude. Our work provides a methodology for performing statistical analysis of the redshift drift effect and deriving its fluctuation power spectra from future large scale surveys., Comment: 19 pages, 5 figures, 3 tables

Topology describes global quantities invariant under continuous deformations, such as the number of elementary excitations at a phase boundary, without detailing specifics. Conversely, differential laws are needed to understand the physical properties of these excitations, such as their localization and spatial behavior. For instance, topology mandates the existence of solitonic zero-energy modes at the domain walls between topologically inequivalent phases in topological insulators and superconductors. However, the spatial dependence of these modes is only known in the idealized (and unrealistic) case of a sharp domain wall. Here, we find the analytical solutions of these zero-modes by assuming a smooth and exponentially-confined domain wall. This allows us to characterize the zero-modes using a few length scales: the domain wall width, the exponential decay length, and oscillation wavelength. These quantities define distinct regimes: featureless modes with "no hair" at sharp domain walls, and nonfeatureless modes at smooth domain walls, respectively, with "short hair", i.e., featureless at long distances, and "long hair", i.e., nonfeatureless at all length scales. We thus establish a universal relation between the bulk excitation gap, decay rate, and oscillation momentum of the zero modes, which quantifies the bulk-boundary correspondence in terms of experimentally measurable physical quantities. Additionally, we reveal an unexpected duality between topological zero modes and Shockley modes, unifying the understanding of topologically-protected and nontopological boundary modes. These findings shed some new light on the localization properties of edge modes in topological insulators and Majorana zero modes in topological superconductors and on the differences and similarities between topological and nontopological zero modes in these systems., Comment: 16 pages, 4 figures

Advancements in artificial intelligence have driven the development of numerous pathology foundation models capable of extracting clinically relevant information. However, there is currently limited literature independently evaluating these foundation models on truly external cohorts and clinically-relevant tasks to uncover adjustments for future improvements. In this study, we benchmarked ten histopathology foundation models on 13 patient cohorts with 6,791 patients and 9,493 slides from lung, colorectal, gastric, and breast cancers. The models were evaluated on weakly-supervised tasks related to biomarkers, morphological properties, and prognostic outcomes. We show that a vision-language foundation model, CONCH, yielded the highest performance in 42% of tasks when compared to vision-only foundation models. The experiments reveal that foundation models trained on distinct cohorts learn complementary features to predict the same label, and can be fused to outperform the current state of the art. Creating an ensemble of complementary foundation models outperformed CONCH in 66% of tasks. Moreover, our findings suggest that data diversity outweighs data volume for foundation models. Our work highlights actionable adjustments to improve pathology foundation models.

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

The lack of transparency in the decision-making processes of deep learning systems presents a significant challenge in modern artificial intelligence (AI), as it impairs users' ability to rely on and verify these systems. To address this challenge, Concept Bottleneck Models (CBMs) have made significant progress by incorporating human-interpretable concepts into deep learning architectures. This approach allows predictions to be traced back to specific concept patterns that users can understand and potentially intervene on. However, existing CBMs' task predictors are not fully interpretable, preventing a thorough analysis and any form of formal verification of their decision-making process prior to deployment, thereby raising significant reliability concerns. To bridge this gap, we introduce Concept-based Memory Reasoner (CMR), a novel CBM designed to provide a human-understandable and provably-verifiable task prediction process. Our approach is to model each task prediction as a neural selection mechanism over a memory of learnable logic rules, followed by a symbolic evaluation of the selected rule. The presence of an explicit memory and the symbolic evaluation allow domain experts to inspect and formally verify the validity of certain global properties of interest for the task prediction process. Experimental results demonstrate that CMR achieves comparable accuracy-interpretability trade-offs to state-of-the-art CBMs, discovers logic rules consistent with ground truths, allows for rule interventions, and allows pre-deployment verification.

Causal opacity denotes the difficulty in understanding the "hidden" causal structure underlying the decisions of deep neural network (DNN) models. This leads to the inability to rely on and verify state-of-the-art DNN-based systems, especially in high-stakes scenarios. For this reason, circumventing causal opacity in DNNs represents a key open challenge at the intersection of deep learning, interpretability, and causality. This work addresses this gap by introducing Causal Concept Graph Models (Causal CGMs), a class of interpretable models whose decision-making process is causally transparent by design. Our experiments show that Causal CGMs can: (i) match the generalisation performance of causally opaque models, (ii) enable human-in-the-loop corrections to mispredicted intermediate reasoning steps, boosting not just downstream accuracy after corrections but also the reliability of the explanations provided for specific instances, and (iii) support the analysis of interventional and counterfactual scenarios, thereby improving the model's causal interpretability and supporting the effective verification of its reliability and fairness.

Neuro-symbolic systems (NeSy), which claim to combine the best of both learning and reasoning capabilities of artificial intelligence, are missing a core property of reasoning systems: Declarativeness. The lack of declarativeness is caused by the functional nature of neural predicates inherited from neural networks. We propose and implement a general framework for fully declarative neural predicates, which hence extends to fully declarative NeSy frameworks. We first show that the declarative extension preserves the learning and reasoning capabilities while being able to answer arbitrary queries while only being trained on a single query type.

Complex non-local behavior makes designing high efficiency and multifunctional metasurfaces a significant challenge. While using libraries of meta-atoms provide a simple and fast implementation methodology, pillar to pillar interaction often imposes performance limitations. On the other extreme, inverse design based on topology optimization leverages non-local coupling to achieve high efficiency, but leads to complex and difficult to fabricate structures. In this paper, we demonstrate numerically and experimentally a shape optimization method that enables high efficiency metasurfaces while providing direct control of the structure complexity. The proposed method provides a path towards manufacturability of inverse-designed high efficiency metasurfaces.

Gravitational lensing is a crucial tool for exploring cosmic phenomena, providing insights into galaxy clustering, dark matter, and dark energy. Given the substantial computational demands of $N$-body simulations, approximate methods like $\texttt{PINOCCHIO}$ and $\texttt{turboGL}$ have been proposed as viable alternatives for simulating lensing probability density functions (PDFs). This paper evaluates these methods and their effectiveness across both weak and strong lensing regimes, with a focus in the context where baryonic effects are negligible. Our comparative analysis reveals that these methods are effective for applications where lensing is mild, such as the majority of sources of electromagnetic and gravitational waves. However, both $\texttt{PINOCCHIO}$ and $\texttt{turboGL}$ break down for large values of convergence and magnification due to their loss of accuracy in capturing small-scale nonlinear matter fields, owing to oversimplified assumptions about internal halo structures and reliance on perturbation theory. $\texttt{PINOCCHIO}$ yields second-to-fourth moments of the lensing PDFs, which are 6-10% smaller than those resulting from $N$-body simulations in regimes where baryonic effects are minimal. These findings aim to inform future studies on gravitational lensing of point sources, which are increasingly relevant with upcoming supernova and gravitational wave datasets., Comment: 16 pages, 12 figures and 4 tables

We report on the detection of X-ray polarisation in the black-hole X-ray binary Swift J1727.8$-$1613 during its dim hard spectral state by the Imaging X-ray Polarimetry Explorer (IXPE). This is the first detection of X-ray polarisation at the transition from the soft to the hard state in an X-ray binary. We find an averaged 2$-$8 keV polarisation degree of (3.3 ${\pm}$ 0.4) % and a corresponding polarisation angle of 3{\deg} ${\pm}$ 4{\deg}, which matches the polarisation detected during the rising stage of the outburst, in September$-$October 2023, within 1${\sigma}$ uncertainty. The observational campaign complements previous studies of this source and enables comparison of the X-ray polarisation properties of a single transient across the X-ray hardness-intensity diagram. The complete recovery of the X-ray polarisation properties, including the energy dependence, came after a dramatic drop in the X-ray polarisation during the soft state. The new IXPE observations in the dim hard state at the reverse transition indicate that the accretion properties, including the geometry of the corona, appear to be strikingly similar to the bright hard state during the outburst rise despite the X-ray luminosities differing by two orders of magnitude., Comment: 8 pages, 6 figures. Accepted for publication in A&A Letters

With its 12 optical filters, the Javalambre-Photometric Local Universe Survey (J-PLUS) provides an unprecedented multicolor view of the local Universe. The third data release (DR3) covers 3,192 deg$^2$ and contains 47.4 million objects. However, the classification algorithms currently implemented in its pipeline are deterministic and based solely on the sources morphology. Our goal is classify the sources identified in the J-PLUS DR3 images into stars, quasi-stellar objects (QSOs), and galaxies. For this task, we present BANNJOS, a machine learning pipeline that uses Bayesian neural networks to provide the probability distribution function (PDF) of the classification. BANNJOS is trained on photometric, astrometric, and morphological data from J-PLUS DR3, Gaia DR3, and CatWISE2020, using over 1.2 million objects with spectroscopic classification from SDSS DR18, LAMOST DR9, DESI EDR, and Gaia DR3. Results are validated using $1.4 10^5$ objects and cross-checked against theoretical model predictions. BANNJOS outperforms all previous classifiers in terms of accuracy, precision, and completeness across the entire magnitude range. It delivers over 95% accuracy for objects brighter than $r = 21.5$ mag, and ~90% accuracy for those up to $r = 22$ mag, where J-PLUS completeness is < 25%. BANNJOS is also the first object classifier to provide the full probability distribution function (PDF) of the classification, enabling precise object selection for high purity or completeness, and for identifying objects with complex features, like active galactic nuclei with resolved host galaxies. BANNJOS has effectively classified J-PLUS sources into around 20 million galaxies, 1 million QSOs, and 26 million stars, with full PDFs for each, which allow for later refinement of the sample. The upcoming J-PAS survey, with its 56 color bands, will further enhance BANNJOS's ability to detail each source's nature., Comment: Submitted to Astronomy and Astrophysics. 29 pages, 23 figures, 4 tables

This paper details our journey in designing and selecting a suitable application sandboxing mechanism for a satellite under development, with a focus on small satellites. Central to our study is the development of selection criteria for sandboxing and assessing its appropriateness for our satellite payload. We also test our approach on two already operational satellites, Suchai and SALSAT, to validate its effectiveness. These experiments highlight the practicality and efficiency of our chosen sandboxing method for real-world space systems. Our results provide insights and highlight the challenges involved in integrating application sandboxing in the space sector., Comment: 8 pages, 5 figures, accepted to SpaceSec Workshop 2024 and to be published as post-conference proceedings with NDSS 2024

Aims. To develop a deep-learning based system for recognition of subclinical atherosclerosis on a plain frontal chest x-ray. Methods and Results. A deep-learning algorithm to predict coronary artery calcium (CAC) score (the AI-CAC model) was developed on 460 chest x-ray (80% training cohort, 20% internal validation cohort) of primary prevention patients (58.4% male, median age 63 [51-74] years) with available paired chest x-ray and chest computed tomography (CT) indicated for any clinical reason and performed within 3 months. The CAC score calculated on chest CT was used as ground truth. The model was validated on an temporally-independent cohort of 90 patients from the same institution (external validation). The diagnostic accuracy of the AI-CAC model assessed by the area under the curve (AUC) was the primary outcome. Overall, median AI-CAC score was 35 (0-388) and 28.9% patients had no AI-CAC. AUC of the AI-CAC model to identify a CAC>0 was 0.90 in the internal validation cohort and 0.77 in the external validation cohort. Sensitivity was consistently above 92% in both cohorts. In the overall cohort (n=540), among patients with AI-CAC=0, a single ASCVD event occurred, after 4.3 years. Patients with AI-CAC>0 had significantly higher Kaplan Meier estimates for ASCVD events (13.5% vs. 3.4%, log-rank=0.013). Conclusion. The AI-CAC model seems to accurately detect subclinical atherosclerosis on chest x-ray with elevated sensitivity, and to predict ASCVD events with elevated negative predictive value. Adoption of the AI-CAC model to refine CV risk stratification or as an opportunistic screening tool requires prospective evaluation., Comment: Submitted to European Heart Journal - Cardiovascular Imaging Added also the additional material 44 pages (30 main paper, 14 additional material), 14 figures (5 main manuscript, 9 additional material)

The cosmological principle posits that the universe is statistically homogeneous and isotropic on large scales, implying all matter shares the same rest frame. This principle suggests that velocity estimates of our motion from various sources should agree with the cosmic microwave background (CMB) dipole's inferred velocity of 370 km/s. Yet, for over two decades, analyses of radio galaxy and quasar catalogs have found velocities at odds with the CMB dipole, with tensions up to 5$\sigma$. In a blind analysis of BOSS and eBOSS spectroscopic data from galaxies and quasars across $0.2

Black-hole X-ray binaries exhibit different spectral and timing properties in different accretion states. The X-ray outburst of a recently discovered and extraordinarily bright source, Swift$~$J1727.8$-$1613, has enabled the first investigation of how the X-ray polarization properties of a source evolve with spectral state. The 2$-$8 keV polarization degree was previously measured by the Imaging X-ray Polarimetry Explorer (IXPE) to be $\approx$ 4% in the hard and hard intermediate states. Here we present new IXPE results taken in the soft state, with the X-ray flux dominated by the thermal accretion-disk emission. We find that the polarization degree has dropped dramatically to $\lesssim$ 1%. This result indicates that the measured X-ray polarization is largely sensitive to the accretion state and the polarization fraction is significantly higher in the hard state when the X-ray emission is dominated by up-scattered radiation in the X-ray corona. The combined polarization measurements in the soft and hard states disfavor a very high or low inclination of the system., Comment: 9 pages, 7 figures

We propose a data-driven methodology to learn a low-dimensional actuation manifold of controlled flows. The starting point is resolving snapshot flow data for a representative ensemble of actuations. Key enablers for the actuation manifold are isometric mapping as encoder and k-nearest neighbour regression as a decoder. This methodology is tested for the fluidic pinball, a cluster of three parallel cylinders perpendicular to the oncoming uniform flow. The centers of these cylinders are the vertices of an equilateral triangle pointing upstream. The flow is manipulated by constant rotation of the cylinders, i.e. described by three actuation parameters. The Reynolds number based on a cylinder diameter is chosen to be 30. The unforced flow yields statistically symmetric unforced periodic shedding represented by a one-dimensional limit cycle. The proposed methodology yields a five-dimensional manifold describing a wide range of dynamics with small representation error. Interestingly, the manifold coordinates automatically unveil physically meaningful parameters. Two of them describe the downstream periodic vortex shedding. The other three ones describe the near-field actuation, i.e. the strength of boat-tailing, the Magnus effect and forward stagnation point. The manifold is shown to be a key enabler for control-oriented flow estimation., Comment: 10 pages, 5 figures

The foundation of two pioneering Usenet newsgroups in astrophysics - still existent today - and some of the main milestones in their history have been tracked from the origins at Princeton University in 1983 to 1994. To this aim, in line with authoritative recommendations from the discipline of web history, different kinds of sources have been retrieved and combined: mainly, online archives of Usenet newsgroups and human sources. The latter have included on one side the kind contribution provided by four astrophysicists, two of which had a role in the maintaining of these newsgroups, through individual interviews; on the other side that from the sixty-seven researchers who answered a purpose-built questionnaire submitted within the Italian National Institute for Astrophysics in May 2023., Comment: provisional draft; 34 pages

Current deep learning models are not designed to simultaneously address three fundamental questions: predict class labels to solve a given classification task (the "What?"), simulate changes in the situation to evaluate how this impacts class predictions (the "How?"), and imagine how the scenario should change to result in different class predictions (the "Why not?"). The inability to answer these questions represents a crucial gap in deploying reliable AI agents, calibrating human trust, and improving human-machine interaction. To bridge this gap, we introduce CounterFactual Concept Bottleneck Models (CF-CBMs), a class of models designed to efficiently address the above queries all at once without the need to run post-hoc searches. Our experimental results demonstrate that CF-CBMs: achieve classification accuracy comparable to black-box models and existing CBMs ("What?"), rely on fewer important concepts leading to simpler explanations ("How?"), and produce interpretable, concept-based counterfactuals ("Why not?"). Additionally, we show that training the counterfactual generator jointly with the CBM leads to two key improvements: (i) it alters the model's decision-making process, making the model rely on fewer important concepts (leading to simpler explanations), and (ii) it significantly increases the causal effect of concept interventions on class predictions, making the model more responsive to these changes.

This study presents a noise-robust closed-loop control strategy for wake flows employing model predictive control. The proposed control framework involves the autonomous offline selection of hyperparameters, eliminating the need for user interaction. To this purpose, Bayesian optimization maximizes the control performance, adapting to external disturbances, plant model inaccuracies, and actuation constraints. The noise robustness of the control is achieved through sensor data smoothing based on local polynomial regression. The plant model can be identified through either theoretical formulation or using existing data-driven techniques. In this work, we leverage the latter approach, which requires minimal user intervention. The self-tuned control strategy is applied to the control of the wake of the fluidic pinball, with the plant model based solely on aerodynamic force measurements. The closed-loop actuation results in two distinct control mechanisms: boat tailing for drag reduction and stagnation point control for lift stabilization. The control strategy proves to be highly effective even in realistic noise scenarios, despite relying on a plant model based on a reduced number of sensors.

A common problem in clinical trials is to test whether the effect of an explanatory variable on a response of interest is similar between two groups, e.g. patient or treatment groups. In this regard, similarity is defined as equivalence up to a pre-specified threshold that denotes an acceptable deviation between the two groups. This issue is typically tackled by assessing if the explanatory variable's effect on the response is similar. This assessment is based on, for example, confidence intervals of differences or a suitable distance between two parametric regression models. Typically, these approaches build on the assumption of a univariate continuous or binary outcome variable. However, multivariate outcomes, especially beyond the case of bivariate binary response, remain underexplored. This paper introduces an approach based on a generalised joint regression framework exploiting the Gaussian copula. Compared to existing methods, our approach accommodates various outcome variable scales, such as continuous, binary, categorical, and ordinal, including mixed outcomes in multi-dimensional spaces. We demonstrate the validity of this approach through a simulation study and an efficacy-toxicity case study, hence highlighting its practical relevance., Comment: 20 pages, 5 figures