The predominant approach for training web navigation agents is to gather human demonstrations for a set of popular websites and hand-written tasks, but it is becoming clear that human data is an inefficient resource. We develop a pipeline to facilitate internet-scale training for agents without laborious human annotations. In the first stage, an LLM annotates 150k sites with agentic tasks. In the next stage, LLM agents complete tasks and produce trajectories. In the final stage, an LLM filters trajectories by judging their success. Language models are powerful data curation tools, identifying harmful content with an accuracy of 97 82.6 are competitive with frontier LLMs as web agents, while being smaller and faster. Our top agent reaches a success rate of 56.9 collection policy Qwen 3 235B, a 235 times larger Llama 4 Maverick, and reaching 94.7 models and data at: https://data-for-agents.github.io.

A better understanding of how age-related bone loss affects the fracture-prone regions of the proximal femur could lead to more informed fracture-prevention strategies. Therefore, the aim of this work was to assess the spatio-temporal distribution of bone deterioration in older men and women with aging. A subset of 305 men (74.87 ± 4.76 years; mean ± SD) and 371 age-matched women (74.84 ± 4.71 years) with no history of fracture was randomly selected from the Age, Gene/Environment Susceptibility-Reykjavik study. Quantitative computed tomography (QCT) scans of the left proximal femur obtained at baseline and at 5.2 ± 0.4 years follow-up were processed to assess local changes in volumetric bone mineral density (vBMD), cortical bone thickness (Ct.Th), and internal bone structure using voxel-based morphometry (VBM), surface-based statistical parametric mapping (surf-SPM), and tensor-based morphometry (TBM). Local parametric changes within each sex and sex differences in these changes were statistically assessed using linear mixed effects models allowing for baseline and time-varying covariates, yielding Student's t-test and p-value statistical maps of the proximal femur. The statistical maps indicated regions with significant parametric changes in each sex and with significant different parametric changes between older men and older women with aging. Older women manifested significantly larger losses in vBMD, (Ct.Th), and structure than older men, and they did so in regions where deficiency in these parameters has been associated with incident hip fracture. Using longitudinal QCT scans of the proximal femur and Computational Anatomy, we provided new insights into the higher fracture rates of the proximal femur in older women compared with men of similar age providing new information on the pathophysiology of osteoporosis.

In this paper, we introduce a new challenging task called Zero-Shot Controllable Image-to-Video Animation, where the goal is to animate an image based on motion trajectories defined by the user, without fine-tuning the base model. Primary challenges include maintaining consistency of background, consistency of object in motion, faithfulness to the user-defined trajectory, and quality of motion animation. We also introduce a novel approach for this task, leveraging diffusion models called IMG2VIDANIM-ZERO (IVA(0)). IVA(0) tackles our controllable Image-to-Video (I2V) task by decomposing it into two subtasks: 'out-of-place' and 'in-place' motion animation. Due to this decomposition, IVA(0) can leverage existing work on layout-conditioned image generation for out-of-place motion generation, and existing text-conditioned video generation methods for in-place motion animation, thus facilitating zero-shot generation. Our model also addresses key challenges for controllable animation, such as Layout Conditioning via Spatio-Temporal Masking to incorporate user guidance and Motion Afterimage Suppression (MAS) scheme to reduce object ghosting during out-of-place animation. Finally, we design a novel controllable I2V benchmark featuring diverse local- and global-level metrics. Results show IVA(0) as a new state-of-the-art, establishing a new standard for the zero-shot controllable I2V task. Our method highlights the simplicity and effectiveness of task decomposition and modularization for this novel task for future studies. Our code and visualizations are available at https://img2vidanim-0.github.io/

Recent advancements in text-to-image generation models have witnessed the success of large-scale diffusion-based generative models. However, exerting control over these models, particularly for structure-conditioned text-to-image generation, remains an open challenge. One straightforward way to achieve control is via fine-tuning, often coming at the cost of efficiency. In this work, we address this challenge by introducing ELR-Diffusion (Efficient Low-rank Diffusion), a method tailored for efficient structure-conditioned image generation. Our innovative approach leverages the low-rank decomposition of model weights, leading to a dramatic reduction in memory cost and model parameters — by up to 58\%, at the same time performing comparably to larger models trained with expansive datasets and more computational resources. At the heart of ELR-Diffusion lies a two-stage training scheme that resorts to the low-rank decomposition and knowledge distillation strategy. To provide a robust assessment of our model, we undertake a thorough comparative analysis in the controllable text-to-image generation domain. We employ a diverse array of evaluation metrics with various conditions, including edge maps, segmentation maps, and image quality measures, offering a holistic view of the model's capabilities. We believe that ELR-Diffusion has the potential to serve as an efficient foundation model for diverse user applications that demand accurate comprehension of inputs containing multiple conditional information.

Physically rearranging objects is an important capability for embodied agents. Visual room rearrangement evaluates an agent's ability to rearrange objects in a room to a desired goal based solely on visual input. We propose a simple yet effective method for this problem: (1) search for and map which objects need to be rearranged, and (2) rearrange each object until the task is complete. Our approach consists of an off-the-shelf semantic segmentation model, voxel-based semantic map, and semantic search policy to efficiently find objects that need to be rearranged. On the AI2-THOR Rearrangement Challenge, our method improves on current state-of-the-art end-to-end reinforcement learning-based methods that learn visual rearrangement policies from 0.53% correct rearrangement to 16.56%, using only 2.7% as many samples from the environment.

Milk and milk products have been known as important for bone health. Can ingestion of milk and milk products lower hip fracture risk for older adults? In this study, older Icelandic adults who were ingesting higher milk had a lower risk of hip fractures. This study describes associations between milk intake and hip fracture risk in older Icelanders. The data indicate that no/low milk consumption is related to greater hip fracture risk. Hip fracture can have a severe effect on the life of older adults. Health authorities recommend milk intake for better bone health. However, previous studies addressing this association have been divergent. This prospective study included 4614 subjects (mean age 76 years) recruited between 2002 and 2006 into the Age, Gene/Environment Susceptibility-Reykjavik (AGES-Reykjavik) study. Information on hip fractures occurring between recruitment and end of follow-up in 2012 was extracted from hospital records. A total of 14

Automatic song writing is a topic of significant practical interest. However, its research is largely hindered by the lack of training data due to copyright concerns and challenged by its creative nature. Most noticeably, prior works often fall short of modeling the cross-modal correlation between melody and lyrics due to limited parallel data, hence generating lyrics that are less singable. Existing works also lack effective mechanisms for content control, a much desired feature for democratizing song creation for people with limited music background. In this work, we propose to generate pleasantly listenable lyrics without training on melody-lyric aligned data. Instead, we design a hierarchical lyric generation framework that disentangles training (based purely on text) from inference (melody-guided text generation). At inference time, we leverage the crucial alignments between melody and lyrics and compile the given melody into constraints to guide the generation process. Evaluation results show that our model can generate high-quality lyrics that are more singable, intelligible, coherent, and in rhyme than strong baselines including those supervised on parallel data.

Household environments are visually diverse. Embodied agents performing Vision-and-Language Navigation (VLN) in the wild must be able to handle this diversity, while also following arbitrary language instructions. Recently, Vision-Language models like CLIP have shown great performance on the task of zero-shot object recognition. In this work, we ask if these models are also capable of zero-shot language grounding. In particular, we utilize CLIP to tackle the novel problem of zero-shot VLN using natural language referring expressions that describe target objects, in contrast to past work that used simple language templates describing object classes. We examine CLIP's capability in making sequential navigational decisions without any dataset-specific finetuning, and study how it influences the path that an agent takes. Our results on the coarse-grained instruction following task of REVERIE demonstrate the navigational capability of CLIP, surpassing the supervised baseline in terms of both success rate (SR) and success weighted by path length (SPL). More importantly, we quantitatively show that our CLIP-based zero-shot approach generalizes better to show consistent performance across environments when compared to SOTA, fully supervised learning approaches when evaluated via Relative Change in Success (RCS).

The risk of a recurrent fragility fracture is high following a first fracture and higher still with more than one prior fracture. This study provides adjustments to FRAX-based fracture probabilities accounting for the number of prior fractures. Prior fractures increase subsequent fracture risk. The aim of this study was to quantify the effect of the number of prior fractures on the 10-year probability of fracture determined with FRAX®. The study used data from the Reykjavik Study fracture register that documented prospectively all fractures at all skeletal sites in a large sample of the population of Iceland. Ten-year probabilities of hip fracture and major osteoporotic fracture (MOF) were determined according to the number of prior osteoporotic fractures over a 20-year interval from the hazards of death and fracture. Fracture probabilities were also computed for a prior osteoporotic fracture irrespective of the number of previous fractures. The probability ratios provided adjustments to conventional FRAX estimates of fracture probability according to the number of prior fractures. Probability ratios to adjust 10-year FRAX probabilities of a hip fracture and MOF increased with the number of prior fractures but decreased with age in both men and women. Probability ratios were similar in men and women and for hip fracture and MOF. Mean probability ratios according to the number of prior fractures for all scenarios were 0.95, 1.08, 1.21 and 1.35, for 1,2, 3 and 4 or more prior fractures, respectively. Thus, a simple rule of thumb is to downward adjust FRAX-based fracture probabilities by 5% in the presence of a single prior fracture and to uplift probabilities by 10, 20 and 30% with a history of 2, 3 and 4 or more prior fractures, respectively. The probability ratios provide adjustments to conventional FRAX estimates of fracture probability according to the number of prior fractures.

Visual navigation through novel environments is an essential skill for intelligent household robots. Reinforcement learning (RL) with dense reward shaping is the standard approach for such visual navigation problems. Dense rewards are used despite the fact that they penalize exploration—a key skill for navigating in new environments—because of the commonly held belief that sparse rewards do not work. In this paper, we present a surprising finding. We demonstrate that combining sparse rewards with self-supervised learning (SSL) not only makes them work, but also outperforms dense rewards, which is the first result of this kind. We apply our approach alongside data augmentation to train a general-purpose agent for image-goal navigation (ImageNav), in which the goal is presented as an image captured from the target location. These techniques improve navigation success from 64% to 72% (+8), path efficiency—measured by success weighted by path length (SPL)—from 0.50 to 0.62 (+0.12), and also lead to a +0.06 absolute improvement in SPL over the state-of-the-art.

INTRODUCTION:Educational attainment is related to improved health and longevity. We investigated the relationship between educational attainment and cardiovascular risk factors, subclinical atherosclerosis, and incidence of coronary artery disease.MATERIAL AND METHODS:The Reykjavik REFINE study is a population-based study recruiting 6616 subjects, 25-69 years of age from the greater Reykjavik area in 2005-2011. Baseline measurements of cardiovascular risk factors were performed, and all participants had a carotid ultrasound examination to detect subclinical atherosclerotic lesions. Clinical follow-up of cardiovascular disease during a ten-year period was performed. Educational attainment was related to clinical outcome measures.RESULTS:The study population comprised of 3251 men and 3365 women. The proportion of the study population with primary school education only was 20.1%, 31.2% had vocational training, 12.3% had high school education and 36.4% were university graduates. Traditional cardiovascular risk factors were generally higher among subjects with primary school education only. Compared to subjects with university education, the odds ratio of having severe atherosclerotic plaque was 1.84 (95% CI 1.40-2.43) among those with primary school education only and 1.49 (95% CI 1.16-1.91) among subjects with vocational training. The subjects with high school or university education were less likely to develop significant cardiovascular disease during the 10-year follow-up period.CONCLUSION:Primary school and vocational training compared to university education are associated with risk factors of atherosclerotic disease, subclinical carotid plaque, and incidence of cardiovascular disease. The reason for this disparity remains to be clarified but socioeconomic inequality related to less educational attainment might be involved.

Abstract Introduction Sex based differences exist in the presentation, outcomes, and management of cardiovascular diseases. Although aortic valve calcium (AVC) score has been postulated to be associated with post-procedural outcomes after transcatheter aortic valve replacement (TAVR), data on the impact of AVC score on procedural outcomes after TAVR based on sex have been scarce. Objective The aim of the present study was to elucidate sex related differences in the prognostic impact of AVC score as a predictor of post-procedural outcomes and mortality in patients with severe aortic stenosis (AS) who underwent TAVR. Methods We retrospectively abstracted the records of 497 patients with severe AS who underwent TAVR between January 2016 and July 2019 at our institution. All patients underwent a non-contrast cardiac CT scan on a Siemens Somatom Definition Flash 128 slice scanner. AVC score using the Agatston method was calculated retrospectively. Primary outcome was 1-year mortality. Patients were divided into two groups: 1) Non-severe AVC score group [women <1200 Agatston unit (AU) and men <2000 AU]; 2) Severe AVC Score group [women >1200 AU and men >2000 AU]. Cox-regression model was used to predict effect of variables on 1-year mortality in male and female patients. Results Among 466 patients included, 268 patients were male and 198 were female (57.5% vs 42.5%). When compared with males, female patients had significantly lower AVC score (p<0.001), aortic valve area (AVA) (p<0.001), obstructive CAD (p<0.001), and history of previous PCI (p<0.004), and CABG (<0.001) but had a significantly higher STS score (6.7±3.2 vs 5.8±3.3 P=0.01). There was no difference between need for permanent pacemaker (PPM) implantation (11% vs 9.6%, p=0.4), major complications (16% vs 15%, p=0.9), stroke (0.8% vs 1.6%, p=0.7), 30-day (3.7% vs 4%, p=0.9), and 1-year mortality (14% vs 12%, p=0.6) between males and females, respectively. Female patients required smaller bio-prosthetic valves compared with males (26±3.2 vs 30±3, p<0.001). Cox regression analysis for female patients showed BMI, hemoglobin level, and AVA independently predicted 1-year mortality, while there was no impact of severe AVC score (>1200 AU) on 1-year mortality in females. Similarly, Cox regression analysis for male patients showed there was no impact of severe AVC score (>2000 AU) on 1-year mortality in males. When males in the severe AVC group were compared with female patients in the severe AVC group, there was no difference in 30 day (4.3% vs 3.3%, p=0.82) and 1-year mortality (14.2% vs 13.3%, p=0.77). Similarly, in the severe AVC group there was no difference between need for PPM implantation (12.8% vs 12.1%, p=0.9), major complications (19.7% vs 15.8%, p=0.4), and stroke (2.2% vs 1%, p=0.6) between males and females, respectively. Conclusion There were no sex differences in the impact of AVC score on mortality and post-procedural outcomes after TAVR. Funding Acknowledgement Type of funding sources: None.

Abstract Introduction Extent of aortic valve (AV) calcium increases as degenerative aortic valve stenosis progresses. Initial studies postulated AV calcium (AVC) score to be a predictor of complications such as need for pacemaker implantation, paravalvular regurgitation and mortality after Trans-catheter Aortic valve Replacement (TAVR). However, evidence regarding the impact of AVC score as a predictor of post-procedural complications and mortality has been conflicting. Objective This study aimed to determine the prognostic impact of AVC as predictor of post-procedural outcomes and mortality in patients with severe aortic stenosis (AS) who underwent TAVR. Methods We retrospectively abstracted the records of 497 patients with severe AS who underwent TAVR between January 2016 and July 2019 at our institution. All patients underwent a non-contrast cardiac CT scan on a Siemens Somatom Definition Flash 128 slice scanner. AVC score using the Agatston method was obtained retrospectively. Patients were divided into two groups: 1) Non-severe AVC score group [women <1200 Agatston unit (AU) and men <2000 AU]; 2) Severe AV Calcium Score group [women >1200 AU and men >2000 AU]. Primary outcome was 1-year mortality. Results Among 466 patients included in the analysis, 352 patients were included severe AVC group while 114 patients were in the non-severe AVC group. Patients in the severe AVC group were older (81±8 vs 79±8 years), were predominantly males (79% vs 71%), and had less diabetes mellitus (31% vs 42%, p=0.02). Post-TAVR AV dimensionless index (0.58±0.13 vs 0.58±0.12, p=0.8) and AV mean gradient (9.5±4.9 vs 9.2±5.4 mmHg) were not statistically different between both groups. There was no statistical difference in the need for PPM (12% vs 7%, p=0.16), post-TAVR stroke (1.5% vs 0%), post-TAVR major complications (17.4% vs 10%, p=0.07), 30 day (3.7% vs 4.4%), and 1-year mortality (13.6 vs 11.4, p=0.6) between both groups. Conclusion Our study report that the severity of AVC score does not have a prognostic impact on major post-procedural outcomes and mortality after TAVR. Funding Acknowledgement Type of funding sources: None.

The risk of a recurrent fragility fracture varies by age and sex, as by site and recency of sentinel fracture. The recency of prior fractures affects subsequent fracture risk. Variable recency may obscure other factors that affect subsequent fracture risk. The aim of this study was to quantify the effect of a sentinel fracture by site, age, and sex where the recency was held constant. The study used data from the Reykjavik Study fracture register that documented prospectively all fractures at all skeletal sites in a large sample of the population of Iceland. Fracture incidence was compared to that of the general population determined at fixed times after a sentinel fracture (humeral, clinical vertebral, forearm, hip, and minor fractures). Outcome fractures comprised a major osteoporotic fracture and hip fracture. Sentinel osteoporotic fractures were identified in 9504 men and women. Of these, 3616 individuals sustained a major osteoporotic fracture as the first subsequent fracture, of whom 1799 sustained a hip fracture. Hazard ratios for prior fracture were consistently higher in men than in women and decreased progressively with age. Hazard ratios varied according to the site of sentinel fracture with higher ratios for hip and vertebral fracture than for humerus, forearm, or minor osteoporotic fracture. The risk of a recurrent fragility fracture varies by age, sex, and site of sentinel fracture when recency is held constant.

Context: Follicle-stimulating hormone (FSH) concentrations increase during the perimenopausal transition and remain high after menopause. Loss of bone mineral density (BMD) and gain of bone marrow adiposity (BMA) and body fat mass also occur during this time. In mice, blocking the action of FSH increases bone mass and decreases fat mass.Objective: To investigate the associations between endogenous FSH levels and BMD, BMA, and body composition in older adults, independent of estradiol and testosterone levels.Design, Setting, and Participants: Older adults from the AGES-Reykjavik Study, an observational cohort study.Main Outcome Measures: Areal BMD, total body fat, and lean mass were measured with dual-energy x-ray absorptiometry. Lumbar vertebral BMA was measured by H-1-magnetic resonance spectroscopy. Volumetric BMD and visceral and subcutaneous adipose tissue (VAT, SAT) areas were measured with quantitative computed tomography. The least squares means procedure was used to determine sex hormone-adjusted associations between quartiles of serum FSH and BMD, BMA, and body composition.Results: In women (N = 238, mean age 81 years), those in the highest FSH quartile, compared with the lowest quartile, had lower adjusted mean spine integral BMD (-8.6%), lower spine compressive strength index (-34.8%), higher BMA (+8.4%), lower weight (-8.4%), lower VAT (-17.6%), lower lean mass (-6.1%), and lower fat mass (-11.9%) (all P < 0.05). In men, FSH level was not associated with any outcome.Conclusions: Older postmenopausal women with higher FSH levels have higher BMA, but lower BMD and lower fat and lean mass, independent of estradiol and testosterone levels. Longitudinal studies are needed to better understand the underlying mechanisms.

The increase in fracture risk associated with a recent fragility fracture is more appropriately captured using a 10-year fracture probability than 2- or 5-year probabilities. The recency of prior fractures affects subsequent fracture risk. The aim of this study was to quantify the effect of a recent sentinel fracture, by site, on the 2-, 5-, and 10-year probability of fracture. The study used data from the Reykjavik Study fracture register that documented prospectively all fractures at all skeletal sites in a large sample of the population of Iceland. Fracture probabilities were determined after a sentinel fracture (humeral, clinical vertebral, forearm and hip fracture) occurring within the previous 2 years and probabilities for a prior osteoporotic fracture irrespective of recency. The probability ratios were used to adjust fracture probabilities over a 2-, 5-, and 10-year time horizon. As expected, probabilities decreased with decreasing time horizon. Probability ratios varied according to age and the site of sentinel fracture. Probability ratios to adjust for a prior fracture within the previous 2 years were higher the shorter the time horizon, but the absolute increases in fracture probabilities were much reduced. Thus, fracture probabilities were substantially lower with time horizons less than 10 years. The 10-year probability of fractures is the appropriate metric to capture the impact of the recency of sentinel fractures. The probability ratios provide adjustments to conventional FRAX estimates of fracture probability for recent sentinel fractures, adjustments which can readily inform clinical decision-making.

Eye movement and strategic placement of the visual field onto the retina, gives animals increased resolution of the scene and suppresses distracting information. This fundamental system has been missing from video understanding with deep networks, typically limited to 224 by 224 pixel content locked to the camera frame. We propose a simple idea, WorldFeatures, where each feature at every layer has a spatial transformation, and the feature map is only transformed as needed. We show that a network built with these WorldFeatures, can be used to model eye movements, such as saccades, fixation, and smooth pursuit, even in a batch setting on pre-recorded video. That is, the network can for example use all 224 by 224 pixels to look at a small detail one moment, and the whole scene the next. We show that typical building blocks, such as convolutions and pooling, can be adapted to support WorldFeatures using available tools. Experiments are presented on the Charades, Olympic Sports, and Caltech-UCSD Birds-200-2011 datasets, exploring action recognition, fine-grained recognition, and video stabilization.

The risk of a recurrent fragility fracture is particularly high immediately following the fracture. This study provides adjustments to FRAX-based fracture probabilities accounting for the site of a recent fracture. The recency of prior fractures affects subsequent fracture risk. The aim of this study was to quantify the effect of a recent sentinel fracture, by site, on the 10-year probability of fracture determined with FRAX. The study used data from the Reykjavik Study fracture register that documented prospectively all fractures at all skeletal sites in a large sample of the population of Iceland. Fracture probabilities were determined after a sentinel fracture (humeral, clinical vertebral, forearm and hip fracture) from the hazards of death and fracture. Fracture probabilities were computed on the one hand for sentinel fractures occurring within the previous 2 years and on the other hand, probabilities for a prior osteoporotic fracture irrespective of recency. The probability ratios provided adjustments to conventional FRAX estimates of fracture probability for recent sentinel fractures. Probability ratios to adjust 10-year FRAX probabilities of a major osteoporotic fracture for recent sentinel fractures were age dependent, decreasing with age in both men and women. Probability ratios varied according to the site of sentinel fracture with higher ratios for hip and vertebral fracture than for humerus or forearm fracture. Probability ratios to adjust 10-year FRAX probabilities of a hip fracture for recent sentinel fractures were also age dependent, decreasing with age in both men and women with the exception of forearm fractures. The probability ratios provide adjustments to conventional FRAX estimates of fracture probability for recent sentinel fractures.