We consider the generation of extreme waves of flexible features in deep-water surface gravity waves, in the context of the fully nonlinear wave model, i.e. Euler equations of the free surface. This is achieved through a recently developed GenWave approach, coupled with the high-order spectral method to solve the Euler equations. Through an iterative refinement procedure, GenWave progressively refines the initial phases associated with a given wave spectrum, which leads to extreme-wave generation at specified design time and location, with their probability of exceedance quantified. In this study, we consider two types of extreme waves, enabled by the flexibility of GenWave. The first type is a spatially and temporally localised rogue wave, and second is a spatially localised extreme-wave group propagating at a constant speed. The two generated extreme-wave profiles are compared with their prototypes – the Peregrine soliton and regular soliton (or bright soliton) in the context of the nonlinear Schrödinger equation (NLS), with reasonable agreement. For the persistent group, comparison with linear evolution and the NLS identifies nonlinear persistence, faster propagation and bound-wave content beyond the leading-order NLS description.

Model-based active flow control requires predictive models that are accurate, stable and fast enough for real-time optimisation. In controlled wake flows, this is often achieved through reduced-order models (ROMs) that first compress high-dimensional velocity snapshots into a latent space and then learn a time-stepping predictor for the dynamics in the latent space. Here, we study how the choice of the spatial encoder affects the predictability of the resulting latent coordinates for wake flows under control inputs. Using two actuated two-dimensional wake configurations, a simplified truck wake and the fluidic pinball, we compare proper orthogonal decomposition (POD) against nonlinear convolutional autoencoders (CAEs) and two types of variational autoencoders for compression, and evaluate several temporal predictors based on long short-term memory networks. The CAEs achieve higher compression efficiency and sharper short-term reconstructions, but they produce a latent dynamics that is more irregular and with broadband spectral content. As a consequence, long-horizon forecasts degrade faster and show a higher probability of catastrophic divergence than POD-based models. The POD yields smoother latent trajectories that are easier to learn and extrapolate, leading to more reliable predictions beyond the short-term regime. These results reveal a clear trade-off between compactness and forecast accuracy, and suggest that the stability of the latent dynamics prediction can outweigh maximal compression. This is particularly relevant for control strategies rooted in forecasts of the dynamics, such as model predictive control and reinforcement learning. The findings provide practical guidance for designing actuation-aware, hardware-feasible predictive ROMs for real-time flow control.

Wall-bounded turbulent flows are chaotic and multiscale, rendering fast prediction at high Reynolds numbers computationally prohibitive in applications such as wind farms. Classical data assimilation is based on repeated solutions of the governing equations and, thus, inherits this cost. Generative models learn the probability distribution of flow states, enabling scalable probabilistic reconstruction. Our generative framework couples a variational autoencoder with a diffusion transformer to generate four-dimensional spatiotemporal samples. Bayesian conditioning enables data assimilation without retraining and allows statistical constraints to be imposed through sampling. The framework is applied to a subdomain of turbulent plane Couette flow, where the corresponding direct numerical simulations in this generation region requires $O(10^6)$ spatial degrees of freedom. Using $O(10)$ latent spatial degrees of freedom, the model achieves a compression ratio of $O(10^5)$, which is one to two orders of magnitude above prior reports. It reproduces single-point statistics up to fourth order and the energy spectra, as well as the intermittency and phase-sensitive structure captured by velocity-increment and $Q$–$R$ statistics. Two assimilation scenarios demonstrate that, when observations are statistically consistent with the prior, conditional diffusion models with the proposed sampling strategy preserve complex turbulent statistics in the posterior. However, enforcing these constraints while preserving physical fidelity and sample diversity introduces an inherent trade-off. Excessive conditioning can distort the learned prior, paralleling limitations of classical ensemble-based data assimilation, where this can likewise degrade the prior covariance. These results highlight both the promise of diffusion models as probabilistic surrogates and the challenges of conditioning them.

The finite span and twist of wind turbine blades induce spanwise flow, along with a tip vortex. Conventional blade element theory, rooted in a two-dimensional assumption, cannot capture these spanwise effects. However, most existing three-dimensional (3-D) corrections neglect intrinsic geometric effects leading to inaccuracies near the tip and in non-rotating cases. To address this, we propose the 3-D geometric effects model incorporating axial and radial velocity factors (3-DGAR model), a physically consistent correction to blade element momentum theory. The model captures two fundamental phenomena: downwash that modifies the local effective angle of attack and tip pressure loss driven by radial velocity transport. A coherent physical pathway from the tip vortex to load reduction is established: the spanwise flow transports energy from the tip element into the tip vortex, reducing suction-side velocity and thereby weakening suction at the equivalent angle of attack. Furthermore, symbolic regression is then used to determine two unknown velocity factors from limited experimental data, serving as closure terms. The model identifies the local aspect ratio as a key parameter governing downwash and reveals an exponential decay of tip-loss effects along the span. Validation across blades of varying geometries and operating conditions confirms accurate spanwise flow correction. The results establish a clear link between the correction factors and the underlying flow physics, thereby connecting aerodynamic predictions to flow mechanisms. Overall, the 3-DGAR model provides a functional mapping from geometric inputs to aerodynamic loads, ensuring broad applicability across diverse blade designs.

Over recent decades, nutritional epidemiology has evolved from nutrient-specific investigations to dietary pattern analysis, complementing traditional dietary assessment methods. With the global aging population and rising chronic disease burden, particularly cancer, identifying modifiable dietary risk factors through holistic dietary assessments is essential. Data-driven dietary pattern analysis methods, including principal component analysis and cluster analysis, have been widely used to derive dietary patterns. However, challenges in cluster quality persist, especially with dietary data. Recent advances in machine learning, including non-linear dimensionality reduction, offer promising strategies to enhance cluster separability in dietary pattern research. This review summarises the evolution of dietary assessment methodologies, outlines the development of dietary pattern analytical approaches and synthesises epidemiological evidence linking clustering-driven dietary patterns to cancer and metabolic disease risk. Furthermore, it highlights methodological gaps and discusses future directions emphasising the integration of machine learning methods for practical application in personalised cancer prevention. Establishing standardised frameworks combining dimensionality reduction and clustering may facilitate the translation of validated dietary patterns into digital platforms for widespread public health impact.

This paper discusses how machine learning could support studying fluid mechanics. We first offer the so-called encoder–decoder perspective that can accommodate a range of problem settings fluid mechanicians often tackle with data-driven techniques. Focusing on nonlinear machine-learning analyses, representative efforts on regression, data compression, prediction and system control are introduced, following the given picture of data-driven fluid mechanics. We then provide a step-by-step guideline for studying fluid flows in a data-driven manner – (i) setting the problem up, (ii) curating the data, (iii) constructing a model, (iv) training a model, (v) assessing the machine-learned outcomes, and (vi) improving the techniques – while discussing what we fluid mechanicians should be mindful of. Importantly, each step necessitates carefully incorporating prior knowledge of flow physics to appropriately design machine-learning studies for complex fluid flows, rather than naively applying black-box techniques. We further perform a case study on data-driven compression of a turbulent wake around an aerofoil, offering example approaches for investigating three-dimensional, unsteady turbulent flows. We hope that this perspective will serve as a practical guide for designing machine-learning studies in fluid mechanics.

Alzheimer’s disease (AD) is the most prominent form of dementia in the USA, with 7·2 million Americans aged 65 years and older, living with AD in 2025. As the US population ages, there is a need for preventative measures to reduce the risk of AD. The apo E ϵ4 (ApoE4) allele is the greatest known genetic risk factor for AD, with heterozygous or homozygous ApoE4 carriers having increased likelihood of developing it. Due to abnormal lipid handling by ApoE4 carriers, precision nutrition may play a major role in delaying the onset of AD in these subjects, with n-3 fatty acids, lipids such as olive oil, vitamins C and E, Mediterranean, ketogenic and high-fibre diets showing promising but still limited evidence for ApoE4 carriers, as most of the current evidence comes from preclinical studies. Additionally, compared with non-carriers, ApoE4 carriers exhibit early gut dysbiosis, an imbalance in the gut microbiota that can disrupt normal digestive, metabolic and immune functions. The potential of probiotics and prebiotics to restore microbial balance and reduce dysbiosis in ApoE4 carriers has also been investigated. This review summarises evidence from observational human cohorts and clinical studies, animal models and in vitro mechanistic studies to examine emerging nutritional and gut-microbiome-based approaches to reduce AD risk in ApoE4 carriers. Future directions include the integration of precision nutrition with multi-omics approaches to allow for scalable and comprehensive analyses of ApoE–diet interactions and utilisation of larger human cohorts to determine the efficacy of nutritional and gut-based AD prevention strategies.

Challenging macrotrends in recent decades have characterised population mental health across the globe, with apparent deterioration in adolescent mental health, as well as underfunded, inequitable, and inefficient psychiatric care systems [1]. Unnecessary delays to treatment are widespread, with systemic difficulties, such as clinician burnout, a dearth of funding, and staff shortages, negatively impacting mental health care across Europe and beyond [2]. With limited clinician time and resources available, new methods for improving mental health care are an urgent priority. Digital technologies, and particularly recent advances in the capabilities of artificial intelligence (AI), offer potential to address some of these challenges, thereby transforming approaches to mental health care.

Effective adaptation and mitigation strategies for climate change require high-resolution projections to inform strategic decision-making. Conventional global climate models, which typically operate at resolutions of 150 to 200 km, lack the capacity to represent essential regional processes. IPSL-AID is a global to regional downscaling tool based on a denoising diffusion probabilistic model designed to address this limitation. Trained on ERA5 reanalysis data, it generates 0.25° resolution fields for temperature, wind, and precipitation using coarse inputs and their spatiotemporal context. It also models probability distributions of fine-scale features to produce plausible scenarios for uncertainty quantification. The model accurately reconstructs statistical distributions, including extreme events, power spectra, and spatial structures. This work highlights the potential of generative diffusion models for efficient climate downscaling with uncertainty estimation. Climate downscaling generally comprises two components—super-resolution and bias correction. This work focuses on the former.

Self-excited thermoacoustic oscillations remain a key obstacle to low-emission combustion, especially in time-aperiodic regimes where quasiperiodic and chaotic dynamics hinder conventional controller design. We experimentally demonstrate data-driven suppression of such aperiodic thermoacoustic oscillations in a flame-driven Rijke tube using genetic programming (GP). In closed-loop form, the GP-discovered control laws are static nonlinear feedback maps that generate, in real time, actuation commands from a sensor signal. Performance is evaluated by a cost function that penalises both the thermoacoustic amplitude (state cost) and the actuation effort (input cost), enabling efficiency comparisons. We benchmark GP closed-loop control (GCC) against GP open-loop control and conventional open-loop harmonic forcing. In the quasiperiodic regime, all three strategies achieve nearly $90\,\%$ amplitude reduction, but the closed-loop strategy is the most efficient. The low-amplitude controlled state is dominated by a preferred (high-gain) mode of the open-flame describing function, and can persist even after control deactivation, indicating basin-of-attraction switching. In the chaotic regime, GCC again outperforms both open-loop strategies, achieving the largest thermoacoustic amplitude reduction with minimal actuation effort. Synchronisation analysis reveals phase trapping, drifting and locking between pressure and heat-release-rate fluctuations, while the 0–1 test and complexity–entropy diagnostics reveal that the GP-suppressed state retains measurable signatures of deterministic chaos. Overall, this study shows that GP control can provide an efficient model-free tool for suppressing aperiodic thermoacoustic oscillations by autonomously exploiting high-receptivity flame pathways with state-dependent actuation.

Adherence to the planetary health diet index (PHDI) is associated with health benefits, but its relationship with depression, particularly the roles of dietary components and symptoms, remains poorly understood. We aimed to (1) compare PHDI component interrelationships between individuals with and without depression; (2) identify PHDI components most predictive of depression using machine learning and (3) identify bridge nodes connecting PHDI components with depressive symptoms. This cross-sectional study analysed data from 25 542 adults in the National Health and Nutrition Examination Survey 2005–2018. The PHDI was calculated from two 24-h dietary recalls. Depression was assessed with the Patient Health Questionnaire-9 with scores ≥10 indicating depression. We employed network analysis to compare dietary structures, an extreme gradient boosting (XGBoost) model interpreted using shapley additive explanations (SHAP) values to identify predictors and bridge network analysis to identify bridge nodes connecting the PHDI and depressive symptom communities. The PHDI network in the non-depressed group was more densely connected than in the depressed group. SHAP analysis of the XGBoost model identified non-starchy vegetables (PHDI2) and added sugars (PHDI14) as the most important predictors, with higher component scores associated with a lower predicted probability of depression. In the bridge network analysis, appetite problems (PHQ5) and depressed mood (PHQ2) emerged as the strongest bridge nodes between the PHDI and depressive symptom communities. A structured, plant-forward dietary pattern was associated with lower depression levels, with appetite problems and depressed mood identified as the bridge nodes between the PHDI and depressive symptom communities.

Archaeological context is a foundational concept, yet its definition remains ambiguous and inconsistent. Traditionally, lithostratigraphic frameworks rooted in geology have long dominated, but recent computational and spatial approaches increasingly redefine stratigraphic units. This shift raises a critical challenge: the lack of clear and consistent terminology for defining stratigraphic units. To explore this issue, we applied natural language processing (NLP) to a broad corpus of Paleolithic literature, focusing on the use of key terms such as “layer,” “level,” “occupation,” and “horizon.” Using SciBERT (Scientific BERT; BERT = Bidirectional Encoder Representations from Transformers) to generate contextual embeddings, we compared their semantic distributions with a control set of distinct terms (lithic, bone, flake, charcoal). The control set produced stable, separable clusters, whereas contextual terms showed extensive semantic overlap and instability, reflecting their interchangeable and inconsistent use. These results underscore that the problem is not only linguistic but theoretical: without clear terminological baselines, new methodological advances risk conflating analytical tools with interpretive outcomes. We argue that terminological precision is essential for comparability across sites and for cumulative research. NLP offers a replicable framework for auditing and clarifying terminology, contributing to greater rigor and interoperability in archaeological interpretation.