While deep learning (DL) methods have expanded the approach to solving state estimation problems in data assimilation (DA), a robust approach requires extending them for uncertainty quantification. In this work, we propose a method to represent location-based uncertainty through random, spatiotemporally coherent displacements in physical space, which are used to sample states in the vicinity of a given reference state. We integrate this strategy into an end-to-end variational DA framework, termed 4DVarNet-LU, and compare its performance with a UNet baseline, demonstrating improved reconstruction skill through the explicit use of prior state information. Trained deterministically in output and stochastically in input, we demonstrate that at inference, we see improvement in skill when using information from initial conditions. Using observation system simulation experiments based on a single-layer quasi-geostrophic model with sparse pseudo-nadir altimetry observations, we evaluate reconstruction mean and ensemble skill. Results show that 4DVarNet-LU consistently outperforms the UNet baseline in terms of root mean square error (RMSE) and continuous ranked probability score (CRPS), with increased robustness to initial condition bias and improved temporal stability. While the method does not propose to account for all different uncertainty sources in DA, it provides a structured way for modeling and training DL models for dealing with location-based uncertainty. Future work may extend this approach to account for additional sources of uncertainty and end-to-end ensemble generation.

Modeling feature interactions in tabular data remains a key challenge in predictive modeling, for example, as used for insurance pricing. This paper proposes the tree-like pairwise interaction network (PIN), a novel neural network architecture that explicitly captures pairwise feature interactions through a shared feed-forward neural network architecture that mimics the structure of decision trees. PIN enables intrinsic interpretability by design, allowing for direct inspection of interaction effects. Moreover, it allows for efficient SHapley’s Additive exPlanation computations because it only involves pairwise interactions. We highlight connections between PIN and established models such as GA$^{2}$Ms, gradient boosting machines, and graph neural networks. Empirical results on the popular French motor insurance dataset show that PIN outperforms both traditional and modern neural network benchmarks in predictive accuracy, while also providing insight into how features interact with each another and how they contribute to the predictions.

In recent years, deep learning has transformed data processing, justifying its use in the current study. These models consist of multiple layers that calibrate weights to optimize accuracy. By mimicking neural processes, these networks facilitate high-performance predictive modeling. This study focused on the application of convolutional neural networks for forecasting Sargassum seaweed arrivals in Guadeloupe, using varied climatic data such as winds, surface currents, satellite images, and observations of seaweed arrivals on the Guadeloupe coast. We tested two different approaches: classification models for forecasting up to 14 days and zone regression models for forecasting up to 14 days, also indicating a probability of algae arrival. The results showed that the classification models achieved a mean accuracy of 94.75%, a mean specificity of 97.42%, and a mean sensitivity of 98.78%. Zone regression models achieved a mean accuracy of around 90%, a mean specificity of around 97%, and a mean sensitivity of around 85%, with a mean MAE of 0.13 and a mean RMSE of 0.18. These models perform better than the decision tree proposed in the state of the art for the same problem.

The cost of high-quality aerodynamics simulations for realistic automotive configurations makes comprehensive design studies unfeasible. Data-driven surrogates (learning from data) are an appealing alternative, and there is no shortage of approaches that target shape-to-aerodynamics predictions. However, there is a fundamental limitation (data insufficiency problem) in this context: owing to the proprietary nature of commercial automotive designs, training datasets are limited to a few freely-available geometries. In a previous work the authors, a strategy to construct datasets for training surrogates was introduced. It enables controlled generation of an arbitrary number of samples, by convex interpolation between a small number of basis geometries. In this work, we extend this strategy by introducing three features, namely size, density, and diversity that characterize more general datasets. These are important to assess how useful is a dataset for a specific prediction task (data for learning). A formal measure of diversity is developed and then, datasets of successively increasing diversity but constant size are constructed. We show that the dataset diversity has an impact on the predictive accuracy of machine learning surrogates. A power-law scaling, $ \varepsilon \hskip0.5em \propto \hskip0.5em {M}^{1/2}\hskip0.1em {m}^{-1/6} $, where $ \varepsilon $ is the prediction error, $ M $ is the diversity, and $ m $ the dataset size, collapses 23 controlled experiments onto a single curve, revealing that diversity dominates size in determining prediction error. The proposed framework allows for more rigorous a priori evaluation of models than is currently possible and can be applied readily to other shape optimization problems.

Research that assesses individual judges’ ability to shape decisions typically focuses on courts that publish separate votes and opinions. Yet, many courts issue per curiam judgments that do not permit public dissent. To overcome this limitation, we use a convolutional neural network (CNN) to model the variation in judges’ expressed preferences from language in aggregated judgments. Specifically, we construct a CNN to analyze the written judgments of judge-rapporteurs and opinions of advocates-general from the Court of Justice of the European Union. Along a pro-/anti-EU dimension, we estimate how judgments differ within (1) each case relative to the advocate-general’s opinion, and (2) each judge-rapporteur, which captures how judges alter their writing across cases. Our results provide novel empirical support for theoretical models of European judicial decision-making: more pro-EU opinions driven by the Court, not the advocate-general or the judge-rapporteur, are associated with larger chambers and stronger external signals of compliance.

Accurate short-term precipitation forecasts predominantly rely on dense weather-radar networks, limiting operational value in places most exposed to climate extremes. We present TUPANN (Transferable and Universal Physics-Aligned Nowcasting Network), a satellite-only model trained on GOES-16 RRQPE. Unlike most deep learning models for nowcasting, TUPANN decomposes the forecast into physically meaningful components: a variational encoder–decoder infers motion and intensity fields from recent imagery under optical-flow supervision, a lead-time-conditioned MaxViT evolves the latent state, and a differentiable advection operator reconstructs future frames. We evaluate TUPANN on both GOES-16 and IMERG data, in up to four distinct climates (Rio de Janeiro, Manaus, Miami, La Paz) at 10–180-min lead times using the CSI and HSS metrics over 4–64 mm/h thresholds. Comparisons against optical-flow, deep learning, and hybrid baselines show that TUPANN achieves the best or second-best skill in most settings, with pronounced gains at higher thresholds. Training in multiple cities further improves performance, while cross-city experiments show modest degradation and occasional gains for rare heavy-rain regimes. The model produces smooth, interpretable motion fields aligned with numerical optical flow and runs in near real time due to the low latency of GOES-16. These results indicate that physically aligned learning can provide nowcasts that are skillful, transferable, and global.

The James Webb Space Telescope (JWST) hosts a non-redundant Aperture Masking Interferometer (AMI) in its Near Infrared Imager and Slitless Spectrograph (NIRISS) instrument, providing the only dedicated interferometric facility aboard – magnitudes more precise than any interferometric experiment previously flown. However, the performance of AMI (and other high resolution approaches such as kernel phase) in recovery of structure at high contrasts has not met design expectations. A major contributing factor has been the presence of uncorrected detector systematics, notably charge migration effects in the H2RG sensor, and insufficiently accurate mask metrology. Here we present Amigo, a data-driven calibration framework and analysis pipeline that forward-models the full JWST AMI system – including its optics, detector physics, and readout electronics – using an end-to-end differentiable architecture implemented in the Jax framework and in particular exploiting the $\partial$Lux optical modelling package. Amigo directly models the generation of up-the-ramp detector reads, using an embedded neural sub-module to capture non-linear charge redistribution effects, enabling the optimal extraction of robust observables, for example kernel amplitudes and phases, while mitigating systematics such as the brighter-fatter effect. We demonstrate Amigo’s capabilities by recovering the AB Dor AC binary from commissioning data with high-precision astrometry, and detecting both HD 206893 B and the inner substellar companion HD 206893 c: a benchmark requiring contrasts approaching 10 mag at separations of only 100 mas. These results exceed outcomes from all published pipelines and re-establish AMI as a viable competitor for imaging at high contrast at the diffraction limit. Amigo is publicly available as open-source software community resource .

Annual bluegrass (Poa annua L.) is an extremely problematic weed in turfgrass, posing a significant challenge for turfgrass management. Injudicious use of herbicides for controlling this weed has led to resistance issues and environmental concerns. Site-specific weed control offers an opportunity to achieve effective weed control with less herbicide use, but it requires the development of a pipeline for weed detection and localization, as well as a path planning algorithm (PPA). To achieve this, unmanned aerial system (UAS)-based RGB imagery of P. annua plants in bermudagrass [Cynodon dactylon (L.) Pers.] turf was collected at different weed growth stages at two locations in Texas: Deer Park and College Station. A convolutional neural network (CNN) You Only Look Once (YOLO11) and a real-time detection transformer (RT-DETR) model were evaluated for weed detection. The results showed that the YOLO11n model achieved the highest F1 score (0.64) and mean average precision (mAP)@0.50 (0.68), while the RT-DETR-x model achieved the lowest F1 score (0.52) and mAP@0.50 (0.51). The GeoTransform function transforms image coordinates into a world coordinate system with centimeter-level accuracy (mean error = 1.5 cm). However, the precision of the transformation depends on the quality of the orthophoto georeferencing. Additionally, the PPA showed a significant reduction (37.7%) in travel distance compared with the original weed model–derived distance. The research highlighted the potential of UAS-based imagery for weed detection and localization in turfgrass. Further improvements are needed to enhance model performance by modifying the model architecture (e.g., input image size, hyperparameters) and evaluating its robustness across different weed growth stages and turfgrass species.

Pater's (2019) target article proposes that neural networks will provide theories of learning that generative grammar lacks. We argue that his enthusiasm is premature since the biases of neural networks are largely unknown, and he disregards decades of work on machine learning and learnability. Learning biases form a two-way street: all learners have biases, and those biases constrain the space of learnable grammars in mathematically measurable ways. Analytical methods from the related fields of computational learning theory and grammatical inference allow one to study language learning, neural networks, and linguistics at an appropriate level of abstraction. The only way to satisfy our hunger and to make progress on the science of language learning is to confront these core issues directly.

The target article (Pater 2019) proposes to use neural networks to model learning within existing grammatical frameworks. This is easier said than done. There is a fundamental gap to be bridged that does not receive attention in the article: how can we use neural networks to examine whether it is possible to learn some linguistic representation (a tree, for example) when, after learning is finished, we cannot even tell if this is the type of representation that has been learned (all we see is a sequence of numbers)? Drawing a correspondence between an abstract linguistic representational system and an opaque parameter vector that can (or perhaps cannot) be seen as an instance of such a representation is an implementational mapping problem. Rather than relying on existing frameworks that propose partial solutions to this problem, such as harmonic grammar, I suggest that fusional research of the kind proposed needs to directly address how to ‘find’ linguistic representations in neural network representations.