Reversal Of Gradient Research Articles

Association between pulse wave velocity and cerebral microbleeds: a systematic review and meta-analysis.

Cerebral microbleeds are associated with events that are among the highest mortality and disability events combined worldwide, as well as with hypertensive vasculopathy. The aim of the present study was to investigate the relationship between a marker of hypertensive vasculopathy, arterial stiffness assessed by pulse wave velocity, and cerebral microbleeds. A systematic review and meta-analysis was performed using PubMed, Scopus, and Web of Science, according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) and Cochrane Collaboration Handbook statements. Data extraction, quality assessment and statistical analyses were performed following pre-established criteria. Twenty-one studies involving 18,436 participants were included. Higher levels of pulse wave velocity were associated with a higher presence of cerebral microbleeds p-OR = 1.26 (95% CI; 1.09-1.45), with considerable heterogeneity; even adjusting for potential confounding variables p-OR = 1.12 (95% CI, 1.05-1.20), with substantial heterogeneity. Only the percentage of women was related to p-OR in the adjusted model. Sensitivity analyses confirmed the robustness of our results. Adjusted models showed publication bias. Higher levels of arterial stiffness are associated with greater presence of cerebral microbleeds. This phenomenon may be caused by damage to the brain under higher blood flow loads, in turn due to age-induced reversal of the stiffness gradient between large and small vessels. As the world's population is undergoing demographic ageing, our results underline the importance of establishing pulse wave velocity as a cardiovascular marker for early screening and delaying the onset of the characteristic signs of both diseases.

Impact of stochastic physics on the representation of atmospheric blocking in EC-Earth3

Abstract. Atmospheric blocking is a synoptic-scale phenomenon that consists in an obstruction of the normal easterly progression of weather patterns in the midlatitudes, leading to persistent atmospheric conditions sometimes associated with extreme weather. State-of-the-art climate models systematically underestimate winter atmospheric-blocking frequency, especially over Europe. This is often attributed to a poor representation of small-scale processes that are fundamental for the onset and maintenance of blocking events. Here, we explore how the implementation of two stochastic parameterizations, namely the stochastically perturbed parameterization tendencies (SPPTs) and the stochastic kinetic energy backscatter (SKEB) schemes, influences the representation of Northern Hemisphere winter blocking in EC-Earth3. We show that the activation of the two stochastic schemes has moderate detrimental effects on blocking representation, when assessed through a gradient reversal index. Using a zonal–blocked flow linear decomposition, we attribute such modification to changes in the mean winter atmospheric circulation, primarily manifested in a strengthening of the midlatitude jet stream and an intensification of the Hadley cell. Ultimately, an analysis of the meridional transport of zonal momentum by stationary and transient eddies reveals that these circulation differences arise from changes in tropical stationary-eddy activity. Our findings reconnect with earlier literature on similar experiments and suggest that the activation of stochastic parameterizations may require a retuning of the model to account for the resulting significant changes in the mean atmospheric circulation.

A diffusion-prepared reduced FOV sequence for prostate MRI near metallic implants.

To enable diffusion weighted imaging in prostate patients with metallic total hip replacements in clinically feasible scan times for prostate cancer screening, and avoid distortion and dropout artifacts present in the conventionally used Echo Planar Imaging (EPI). A reduced field of view (FOV) diffusion-prepared sequence that is robust to the B inhomogeneities produced by total hip replacements was achieved using high radiofrequency (RF) bandwidth pulses and manipulation for stimulated echo pathways. The reduced FOV along the A/P direction was obtained using slice-select gradient reversal, and the prepared magnetization was imaged with a three-dimensional RF-spoiled gradient echo readout. The sequence was validated in phantom experiments, in vivo in healthy volunteers with and without total hip replacements, and in vivo in patients undergoing a standard MRI prostate exam. The proposed sequence is robust to shading and distortion artifacts that are encountered by standard diffusion-weighted EPI in the presence of moderate off-resonance. Apparent diffusion coefficient estimates obtained by the proposed sequence were comparable to those obtained with diffusion-weighted EPI. Acquisition of distortionless diffusion weighted images of the prostate is feasible in patients with total hip replacements on conventional, whole-body 3T MRI, using a b-value of 800 and nominal resolution of 1.7 1.7 4 mm3 in scan times of 6 min.

Lysis of human erythrocytes due to Piezo1-dependent cytosolic calcium overload as a mechanism of circulatory removal

Hematopoietic stem cells surrender organelles during differentiation, leaving mature red blood cells (RBC) devoid of transcriptional machinery and mitochondria. The resultant absence of cellular repair capacity limits RBC circulatory longevity, and old cells are removed from circulation. The specific age-dependent alterations required for this apparently targeted removal of RBC, however, remain elusive. Here, we assessed the function of Piezo1, a stretch-activated transmembrane cation channel, within subpopulations of RBC isolated based on physical properties associated with aging. We subsequently investigated the potential role of Piezo1 in RBC removal, using pharmacological and mechanobiological approaches. Dense (old) RBC were separated from whole blood using differential density centrifugation. Tolerance of RBC to mechanical forces within the physiological range was assessed on single-cell and cell population levels. Expression and function of Piezo1 were investigated in separated RBC populations by monitoring accumulation of cytosolic Ca2+ and changes in cell morphology in response to pharmacological Piezo1 stimulation and in response to physical forces. Despite decreased Piezo1 activity with increasing cell age, tolerance to prolonged Piezo1 stimulation declined sharply in older RBC, precipitating lysis. Cell lysis was immediately preceded by an acute reversal of density. We propose a Piezo1-dependent mechanism by which RBC may be removed from circulation: Upon adherence of these RBC to other tissues, they are uniquely exposed to prolonged mechanical forces. The resultant sustained activation of Piezo1 leads to a net influx of Ca2+, overpowering the Ca2+-removal capacity of specifically old RBC, which leads to reversal of ion gradients, dysregulated cell hydration, and ultimately osmotic lysis.

Adversarial Training Based Domain Adaptation of Skin Cancer Images.

Skin lesion datasets used in the research are highly imbalanced; Generative Adversarial Networks can generate synthetic skin lesion images to solve the class imbalance problem, but it can result in bias and domain shift. Domain shifts in skin lesion datasets can also occur if different instruments or imaging resolutions are used to capture skin lesion images. The deep learning models may not perform well in the presence of bias and domain shift in skin lesion datasets. This work presents a domain adaptation algorithm-based methodology for mitigating the effects of domain shift and bias in skin lesion datasets. Six experiments were performed using two different domain adaptation architectures. The domain adversarial neural network with two gradient reversal layers and VGG13 as a feature extractor achieved the highest accuracy and F1 score of 0.7567 and 0.75, respectively, representing an 18.47% improvement in accuracy over the baseline model.

The changing educational and social class gradients in union dissolution: Evidence from a latecomer of the Second Demographic Transition

Most studies on the changing socioeconomic gradient of divorce have operationalized individuals’ socioeconomic status (SES) through education, often neglecting social class differences. Education may proxy cultural and cognitive skills, whereas social class could more accurately capture economic means. Additionally, existing research has predominantly focused on women and marital dissolutions. This study addresses these oversights by analyzing the educational and social class gradients of both marriage and cohabitation dissolutions among men and women in Italy—a latecomer to the Second Demographic Transition. We used non-proportional hazard models to estimate survival curves and union dissolution probabilities stratified by education, social class, and cohort. Our findings reveal a vanishing socioeconomic gradient of marital dissolution among women and a reversal from positive to negative among men across cohorts. These results challenge the conventional view that men’s higher SES always stabilizes unions and support Goode’s hypothesis on the reversal of the socioeconomic gradient of divorce for both genders. No clear SES gradient was found for cohabiting unions. Overall, the study demonstrates the significant predictive power of social class for marital dissolutions, even when controlling for education, emphasizing the need to consider both measures of SES to comprehensively account for different underlying mechanisms.

Unsupervised Transfer Learning Across Different Data Modalities for Bearing's Speed Identification

Recent advancements in transfer learning have revolutionized predictive maintenance, enabling cross-domain generalization for components with varying characteristics and operating under different conditions. While traditional transfer learning approaches require labeled data in both source and target domains, unsupervised transfer learning strives for a more cost-effective alternative for which only labels are available in the source domain. This study investigates adversarial transfer learning between two different sensor modalities: vibration and acoustic. The goal is to enable bearing monitoring using microphones, which are, in general terms, cheaper and easier to deploy than vibration sensors; and without the need to label data in the target domain. The research goal is to identify the operating speed of a bearing testbed. The source domain data correspond to vibration measurements taken from an attached sensor, while the target domain uses a microphone array at distance. Artificial Neural Networks are used as the base architecture. Transferability is assessed with two unsupervised adversarial learning techniques: gradient reversal and deep correlation alignment. Their performance is compared to traditional supervised transfer learning via fine-tuning. Experimental results demonstrate that gradient reversal outperforms deep correlation alignment and is able to achieve results similar to those obtained with supervised transfer learning. These findings highlight the feasibility of speed identification using a microphone array and establish a baseline for future condition monitoring research with such sensors. Conclusion. These findings highlight the feasibility of speed identification using a microphone array and establish a baseline for future condition monitoring research with such sensors.

Objective and neutral summarization of customer reviews

Opinion mining aims to detect and extract relevant information from large quantity of customer reviews. Automatic opinion summarization then seeks to create a consensual point of view often oriented toward the main sentiment of clients to render their experience. Although factual information is valuable for companies to understand what works or not in their products, summarization approaches that convey objectivity, and constructive feedback from customer reviews have yet to be explored. We propose an adversarial multi-task learning model for document summarization to address this new issue. Our algorithm combines an autoencoder for document summarization with a gradient reversal layer to learn independent representations of subjective and sentiment-based material. We assess and compare our method on the Amazon product review dataset where we introduce an original evaluation dataset for objective summarization. We further completed the analysis with neutrality and objectivity metrics. This study demonstrates that the generated summaries carry out relevant and objective content but also emphasize the importance of various processes and layers in multi-task learners to control the information effectively.

Particulate black carbon mass concentrations and the episodic source identification driven by atmospheric blocking effects in Astana, Kazakhstan

Black carbon (BC) is a component of fine particulate matter (PM2.5) that is a key contributor to adverse human health effects and climate forcing. To date, BC mass concentrations and possible sources in Kazakhstan have not been studied. Thus, understanding the temporal variations of BC for a large developing region with a complex climate is useful. In this study, measurements of fine particulate BC mass concentrations in Astana were made from June 2020 to October 2021 by measuring light absorption of PM2.5 on filters. The mean BC concentration was 2.56 ± 1.29 μg m−3 with maximum and minimum monthly mean BC concentrations being 4.56 ± 2.03 μg m−3 and 1.12 ± 0.42 μg m−3 in January 2021 and June 2020, respectively. Temporal analyses of BC, SO2, PM10, NOx, CO, meteorological and atmospheric stability parameters were performed. Aggregated pollutant ‘episodic loadings’ during the heating and non-heating periods were identified. Their relationships with blocking anticyclones and cyclones were investigated by examining the reversal of meridional gradients at 500 hPa geopotential height (GPH) maps and identifying Omega (Ω) and Rex blocking types. Astana has some of the highest BC concentrations of cities worldwide. Seasonal BC source location identification using Conditional Bivariate Probability Function (CBPF) analysis implicated combined heat and power (CHP) plant emissions as the major BC source in Astana. Significant increases in BC concentrations were observed during the cold season due to numerous sources, generally poorer atmospheric dispersion and blocking events. The Concentration Weighted Trajectory (CWT) analysis results showed that the distribution of the 75th percentile of BC during episodic periods actively controlled by blockings exceeding than the entire measurement period, which may reflect cross-border transport and adjacent countries.

Edge preserving infrared and visible image fusion with three layer decomposition based on multi-level co-occurrence filtering

By merging different portraits of a particular scene, image fusion attempts to create a blended image that combines details from all the images. Infrared (IR) and visible image fusion can be accomplished in a variety of ways, including recent deep-learning-based techniques. However, edge-preserving filter (EPF) based fusion works well since it retains all the information from both images. Local filtering-based techniques, on the other hand, limit the fusion performance by introducing multiple gradient reversal artifacts and halos. This work presents an advanced IR and visible image fusion approach depending on three-level decomposition using multi-level co-occurrence filtering, which aims to overcome the common shortfalls such as halo effects seen in existing EPF based fusion. The reference images are decomposed in to base layer, small-scale layers and large-scale layers using multi-level co-occurrence filtering (MLCoF). Since most of the low frequency details are contained in the base layer, the conventional merging strategy by averaging is replaced with novel foreground information map (FIM) based fusion strategy. Small-scale layers are combined by applying max-absolute fusion strategy. A novel weight-map guided edge preserving fusion strategy is put forward for the integration of large-scale layers. Later, fused image is generated by the superposition of these different layers. Subjective visual and objective quantitative analysis shows that the suggested technique attains more notable performance in contrast with other modern fusion methods including many deep-learning techniques. In terms of visual perspective view, the results produced by the proposed approach are superior and include all details from both images. Additionally, it produces outcomes free of gradient reversal and halo artifacts.

Meta-learning based infrared ship object detection model for generalization to unknown domains

Infrared images exhibit considerable variations in probability distributions, stemming from the utilization of distinct infrared sensors and the influence of diverse environmental conditions. The variations pose great challenges for deep learning models to detect ship objects and adapt to unseen maritime environments. To address the domain shift problem, we propose an end-to-end infrared ship object detection model based on meta-learning neural network to improve domain adaptation for target domain where data is not available at training phase. Different from existing domain generalization methods, the novelty of our model lies in the effective exploitation of meta-learning and domain adaptation, ensuring that the extracted domain-independent features are meaningful and domain-invariant at the semantic level. Firstly, a double gradient-based meta-learning algorithm is designed to solve the common optimal descent direction between different domains through two gradient updates in the inner and outer loops. The algorithm enables extraction of domain-invariant features from the pseudo-source and pseudo-target domain data. Secondly, a domain discriminator with dynamic-weighted gradient reversal layer (DWGRL) is designed to accurately classify domain-invariant features and provide additional global supervision information. Finally, a multi-scale feature aggregation method is proposed to improve the extraction of multi-scale domain-invariant features. It can effectively fuse local features at different scales and global features of targets. Extensive experimental results conducted in real nighttime water surface scenes demonstrate that the proposed model achieves very high detection accuracy on target domain data, even no target domain data was used during the training phase. Compared to the existing methods, our method not only improves the detection accuracy of infrared ships by 18%, but also exhibits the smallest standard deviation with a value of 0.93, indicating its superior generalization performance.

Domain generalization of chemical process fault diagnosis by maximizing domain feature distribution alignment

Fault detection and diagnosis (FDD) is crucial for ensuring safety and quality control in chemical industry. But data-driven FDD methods are usually trained with samples under certain operating conditions or data domains. This leads to their performance degradation caused by domain shifts in practice. To address this, a new FDD network, termed the DFDA model, has been designed to achieve domain generalization. Distribution of features extracted from fault samples is progressively aligned by supervised contrastive learning, learning vector quantization and gradient reversal adversarial learning in the DFDA model. So that FDD using such aligned features can potentially offer enhanced domain generalization capabilities. The performance of the DFDA model is benchmarked against other state-of-the-art FDD frameworks using data from various domains/modes of the Tennessee Eastman Process and an actual industrial process. The experimental outcomes indicate that the DFDA model consistently outperforms others in five distinct tasks, achieving an average Fault Detection Rate (FDR) that is 23% higher than that of ViT and 16% higher than CausalViT.

Socioeconomic Position, Pre-Obesity and Obesity in Latin American Cities: A Systematic Review.

Currently the socioeconomic gradient of obesity it is not well understood in the urban population in Latin American. This study reviewed the literature assessing associations between pre-obesity, obesity, and socioeconomic position (SEP) in adults living in urban areas in Latin American countries. PubMed and SciELO databases were used. Data extraction was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We extracted data on the association between SEP (e.g., education, income), pre-obesity (body mass index [BMI] ≥ 25 and < 30 kg/m2) and obesity (BMI ≥ 30 kg/m2). Relative differences between low and high SEP groups were assessed and defined a priori as significant at p < 0.05. Thirty-one studies met our inclusion criteria and most were conducted in Brazil and Mexico (22 and 3 studies, respectively). One study presented nonsignificant associations. Forty-seven percent of associations between education or income and pre-obesity were negative. Regarding obesity, 80 percent were negative and 20 percent positive. Most negative associations were found in women while in men they varied depending on the indicator used. Pre-obesity and obesity by SEP did not follow the same pattern, revealing a reversal of the obesity social gradient by SEP, especially for women in Latin America, highlighting the need for articulated policies that target structural and agentic interventions.

Temporal variability and site specificity of thermomechanical weathering in a temperate climate

Thermomechanical processes caused by short- and long-term temperature fluctuations are a prevalent weathering mechanism on exposed rock walls. While many authors have explored the potential for thermomechanical weathering in alpine and polar regions, few have examined the effects of seasonality on weathering in temperate climates. This is pertinent as seasonal climatic conditions may influence both short-term temperature oscillations which produce incipient fractures and diurnal-to annual-scale cycles which propagate pre-existing fractures via thermal fatigue. In this study, three rock outcrops located along the Niagara Escarpment in Hamilton, Canada were monitored to examine changes in the thermal regime at the rock surface and within pre-existing fractures over a 1-year period. Temperature was sampled in 1-min intervals, providing data at a fine temporal resolution. Our unique dataset demonstrates that the rock surface and fracture experience minute-scale temperature oscillations which magnify over time. Longer-term temperature cycles during the year are superimposed upon minute- and diurnal-scale fluctuations which likely augment weathering potential. This produces considerable thermal stress over the year which we estimate to be on the order of 18 GPa at the rock surface and 8 GPa in fractures. We also observed diurnal reversals of the temperature gradient between the rock surface and fracture which may further amplify crack propagation. Seasonality and site-specific characteristics interact to modify different components of the rockwall thermal regime. Vegetation shading has seasonal and diurnal-scale impacts on the temperature gradient between the surface and fracture, and the amplitude of daily warming and cooling cycles. Aspect has a stronger influence on minute-scale temperature oscillations. Estimates of diurnal thermal stress indicate that the thermomechanical weathering potential is seasonally variable, but highest in the spring. Our findings demonstrate that in a temperate climate, rockwall thermal regimes experience variability across the gradient of temporal scale with strong seasonal effects.

AdvMix: Adversarial Mixing Strategy for Unsupervised Domain Adaptive Object Detection

Recent object detection networks suffer from performance degradation when training data and test data are distinct in image styles and content distributions. In this paper, we propose a domain adaptive method, Adversarial Mixing (AdvMix), where the label-rich source domain and unlabeled target domain are jointly trained by the adversarial feature alignment and a self-training strategy. To diminish the style gap, we design the Adversarial Gradient Reversal Layer (AdvGRL), containing a global-level domain discriminator to align the domain features by gradient reversal, and an adversarial weight mapping function to enhance the stability of domain-invariant features by hard example mining. To eliminate the content gap, we introduce a region mixing self-supervised training strategy where a region of the target image with the highest confidence is selected to merge with the source image, and the synthesis image is self-supervised by the consistency loss. To improve the reliability of self-training, we propose a strict confidence metric combining both object and bounding box uncertainty. Extensive experiments conducted on three benchmarks demonstrate that AdvMix achieves prominent performance in terms of detection accuracy, surpassing existing domain adaptive methods by nearly 5% mAP.

Towards Robust Learning with Noisy and Pseudo Labels for Text Classification

Unlike Positive Training (PT), Negative Training (NT) is an indirect learning technique that trains the model on a combination of clean and noisy data using complementary labels, which are randomly generated from the label space except for the actual label. Although clean samples have identical distributions to the test samples, they are treated with the same level of uncertainty as noisy samples because of the complementary labeling of NT. Consequently, their contribution to the overall performance is relatively lower. We propose a Learning with Noisy and Pseudo Label (LNPL) framework, which jointly trains the model using PT and NT on clean and noisy data, respectively. We aim to enable direct learning on clean samples while leveraging the robustness of NT against noise in a unified framework. To mitigate the abundance of noisy instances, we leverage a gradient reversal layer at the top of LNPL as a regularization term to mislead the recognition of the source of the instance (e.g., clean or noisy). Moreover, we introduce a self-training LNPL that performs a semi-supervised text classification task as a learning with noisy pseudo-label problem. Extensive experiments on various textual benchmark datasets demonstrate that LNPL is robust and consistently outperforms the alternatives. The code is available on GitHub.1

Noise-robust voice conversion using adversarial training with multi-feature decoupling

Most existing voice conversion methods focus primarily on separating speech content from speaker information while overlooking the decoupling of pitch information. Additionally, the quality of converted speech significantly degrades when the speech of the target speaker is contaminated by noises. To address these issues, this paper proposes a noise-robust voice conversion model with multi-feature decoupling based on adversarial training. The proposed framework utilizes three distinct encoders to encode speech content, speaker identity, and pitch information independently, which aims to enhance the performance of decoupling by minimizing their mutual information and reduce the correlations between feature vectors. Moreover, a gradient reversal layer and a noise decoupling discriminator are incorporated into the framework, which extracts noise-resistant speaker representations and content representations through adversarial training to facilitate the synthesis of high-quality speech. In order to optimize the learning process, a training strategy is developed which involves alternating between clean and noisy data during the training of the encoder. This strategy effectively guides and expedites the convergence of the model. Experimental results demonstrate that compared to the state-of-the-art baselines of noise-robust voice conversion, the proposed model achieves improvements around 0.31 and 0.39 in terms of speech naturalness and speaker similarity evaluation metrics, respectively.

An unsupervised subdomain adversarial network for remaining useful life estimation under various conditions

AbstractDomain adaptation is a transfer learning method that is widely applied for remaining useful life (RUL) prediction under different operating conditions. However, current RUL prediction methods based on domain adaptation only consider minimizing degradation feature differences, while ignoring the correlation between features and labels, which may result in significant errors in the mapping relationship between the two learned by the model, and thus reduce the accuracy of prediction. To address this issue, this paper proposes an unsupervised subdomain adversarial network (USDAN) for RUL estimation under various conditions. On the basis of traditional domain adversarial adaptation networks, the network uses temporal convolutional networks to construct the feature extractor to learn the long‐term dependencies between degradation data, and a multilinear conditioning scheme is designed to combine the features and subdomain labels as input features to the domain discriminator. Finally, with backpropagation and gradient reversal, the feature extractor is prompted to generate cross‐domain invariant features that are correlated with the labels, improving the prediction accuracy and enhancing the transferability of the network. In the experimental section, the public turbofan engine degradation dataset is used to validate the effectiveness of the network, and the comparison with other approaches demonstrates its advantages.

Attention guided multi-wavelet adversarial network for cross domain fault diagnosis

Adversarial transfer learning is an effective method for diagnosing rotating machinery faults. However, it does not extract features completely and underutilizes unlabeled target data. Hence, an attention-guided multi-wavelet adversarial network (AMWAN) based on the hybrid metric strategy of a multi-wavelet convolutional feature extractor is proposed. First, hidden fault features in the input data are extracted via multi-wavelet convolution and an attention mechanism. Subsequently, the extracted features are input into the domain discriminator and label classifier. A gradient reversal layer is introduced to map and mix the source- and target- domain samples for feature extraction. Specifically, the k-nearest neighbor is adopted to construct a label propagation model in the source-domain feature layer to obtain pseudo-labels for the target-domain feature data. This extends the labeling information for the target-domain feature data. A hybrid metric based on the maximum mean discrepancy distance and feature cosine metric strategy is introduced in the feature layer for adversarial training, which can significantly enhance the domain adaptive effect and improve the cross-domain fault diagnosis performance of the AMWAN. Experimental validation based on two sets of planetary gearbox datasets indicate that the AMWAN exhibits outstanding fault diagnosis performance under both constant- and variable-speed and -load conditions, i.e., it achieves an average accuracy rate of up to 96.85 %, thus surpassing other methods.

Childhood circumstances, social mobility and the obesity transition: Evidence from South Africa

The distribution of obesity tends to shift from rich to poor individuals as countries develop, in a process of shifting sociodemographic patterns of obesity that has been called the ‘obesity transition’. This change tends to happen with economic development, but little is known about the specific mechanisms that drive the change. We propose that improvements in childhood circumstances with economic development may be one of the drivers of the obesity transition. We explore whether the social gradient in body weight differs by childhood socioeconomic status (SES), proxied by the respondent’s mother having Grade 12, using South Africa’s nationally representative panel National Income Dynamics Study. In support of our hypothesis, we find that the social gradient in body weight is less positive for adults who had a high childhood SES, and already appears to have reversed among high-SES women who also had a high childhood SES. Upward social mobility over an individual’s life course or across a single generation is associated with higher body weight compared to a stable high SES. But a high SES sustained in childhood and adulthood – or across more than one generation – may decrease adult obesity risk, and result in a reversal of the social gradient in body weight. Random effects within-between models show that the social gradient in body weight and its interaction with childhood SES are driven more by differences in income between individuals than by short-run changes in income within individuals, again suggesting that the obesity transition is driven by long-run changes rather than by very short-run changes. Our results are broadly robust to using several alternative measures of body weight, childhood SES and adult SES. Our results are consistent with the hypothesis that widespread improvements in childhood circumstances and nutrition with economic development may contribute to the shift to later stages of the obesity transition.