Explaining the output of a deep network remains a challenge. In the case of an image classifier, one type of explanation is to identify pixels that strongly influence the final decision. A starting point for this strategy is the gradient of the class score function with respect to the input image. This gradient can be interpreted as a sensitivity map, and there are several techniques that elaborate on this basic idea. This paper makes two contributions: it introduces SmoothGrad, a simple method that can help visually sharpen gradient-based sensitivity maps, and it discusses lessons in the visualization of these maps. We publish the code for our experiments and a website with our results.
We present ControlNet, a neural network architecture to add spatial conditioning controls to large, pretrained text-to-image diffusion models. ControlNet locks the production-ready large diffusion models, and reuses their deep and robust encoding layers pretrained with billions of images as a strong backbone to learn a diverse set of conditional controls. The neural architecture is connected with "zero convolutions" (zero-initialized convolution layers) that progressively grow the parameters from zero and ensure that no harmful noise could affect the finetuning. We test various conditioning controls, e.g., edges, depth, segmentation, human pose, etc., with Stable Diffusion, using single or multiple conditions, with or without prompts. We show that the training of ControlNets is robust with small (<50k) and large (>1m) datasets. Extensive results show that ControlNet may facilitate wider applications to control image diffusion models.
Identification of key factors associated with the risk of developing cardiovascular disease and quantification of this risk using multivariable prediction algorithms are among the major advances made in preventive cardiology and cardiovascular epidemiology in the 20th century. The ongoing discovery of new risk markers by scientists presents opportunities and challenges for statisticians and clinicians to evaluate these biomarkers and to develop new risk formulations that incorporate them. One of the key questions is how best to assess and quantify the improvement in risk prediction offered by these new models. Demonstration of a statistically significant association of a new biomarker with cardiovascular risk is not enough. Some researchers have advanced that the improvement in the area under the receiver-operating-characteristic curve (AUC) should be the main criterion, whereas others argue that better measures of performance of prediction models are needed. In this paper, we address this question by introducing two new measures, one based on integrated sensitivity and specificity and the other on reclassification tables. These new measures offer incremental information over the AUC. We discuss the properties of these new measures and contrast them with the AUC. We also develop simple asymptotic tests of significance. We illustrate the use of these measures with an example from the Framingham Heart Study. We propose that scientists consider these types of measures in addition to the AUC when assessing the performance of newer biomarkers.
ABSTRACT We show that analysts from sell‐side firms generally recommend “glamour” (i.e., positive momentum, high growth, high volume, and relatively expensive) stocks. Naïve adherence to these recommendations can be costly, because the level of the consensus recommendation adds value only among stocks with favorable quantitative characteristics (i.e., value stocks and positive momentum stocks). In fact, among stocks with unfavorable quantitative characteristics, higher consensus recommendations are associated with worse subsequent returns. In contrast, we find that the quarterly change in consensus recommendations is a robust return predictor that appears to contain information orthogonal to a large range of other predictive variables.
BACKGROUND: Statin therapy reduces low-density lipoprotein (LDL) cholesterol levels and the risk of cardiovascular events, but whether the addition of ezetimibe, a nonstatin drug that reduces intestinal cholesterol absorption, can reduce the rate of cardiovascular events further is not known. METHODS: We conducted a double-blind, randomized trial involving 18,144 patients who had been hospitalized for an acute coronary syndrome within the preceding 10 days and had LDL cholesterol levels of 50 to 100 mg per deciliter (1.3 to 2.6 mmol per liter) if they were receiving lipid-lowering therapy or 50 to 125 mg per deciliter (1.3 to 3.2 mmol per liter) if they were not receiving lipid-lowering therapy. The combination of simvastatin (40 mg) and ezetimibe (10 mg) (simvastatin-ezetimibe) was compared with simvastatin (40 mg) and placebo (simvastatin monotherapy). The primary end point was a composite of cardiovascular death, nonfatal myocardial infarction, unstable angina requiring rehospitalization, coronary revascularization (≥30 days after randomization), or nonfatal stroke. The median follow-up was 6 years. RESULTS: The median time-weighted average LDL cholesterol level during the study was 53.7 mg per deciliter (1.4 mmol per liter) in the simvastatin-ezetimibe group, as compared with 69.5 mg per deciliter (1.8 mmol per liter) in the simvastatin-monotherapy group (P<0.001). The Kaplan-Meier event rate for the primary end point at 7 years was 32.7% in the simvastatin-ezetimibe group, as compared with 34.7% in the simvastatin-monotherapy group (absolute risk difference, 2.0 percentage points; hazard ratio, 0.936; 95% confidence interval, 0.89 to 0.99; P=0.016). Rates of prespecified muscle, gallbladder, and hepatic adverse effects and cancer were similar in the two groups. CONCLUSIONS: When added to statin therapy, ezetimibe resulted in incremental lowering of LDL cholesterol levels and improved cardiovascular outcomes. Moreover, lowering LDL cholesterol to levels below previous targets provided additional benefit. (Funded by Merck; IMPROVE-IT ClinicalTrials.gov number, NCT00202878.).
Adding It Up explores how students in pre-K through 8th grade learn mathematics and recommends how teaching, curricula, and teacher education should change to improve mathematics learning during these critical years. The committee identifies five interdependent components of mathematical proficiency and describes how students develop this proficiency. With examples and illustrations, the book presents a portrait of mathematics learning: * Research findings on what children know about numbers by the time they arrive in pre-K and the implications for mathematics instruction. * Details on the processes by which students acquire mathematical proficiency with whole numbers, rational numbers, and integers, as well as beginning algebra, geometry, measurement, and probability and statistics. The committee discusses what is known from research about teaching for mathematics proficiency, focusing on the interactions between teachers and students around educational materials and how teachers develop proficiency in teaching mathematics.
This report identifies twelve building block chemicals that can be produced from sugars via biological or chemical conversions. The twelve building blocks can be subsequently converted to a number of high-value bio-based chemicals or materials. Building block chemicals, as considered for this analysis, are molecules with multiple functional groups that possess the potential to be transformed into new families of useful molecules. The twelve sugar-based building blocks are 1,4-diacids (succinic, fumaric and malic), 2,5-furan dicarboxylic acid, 3-hydroxy propionic acid, aspartic acid, glucaric acid, glutamic acid, itaconic acid, levulinic acid, 3-hydroxybutyrolactone, glycerol, sorbitol, and xylitol/arabinitol.
The System Usability Scale (SUS) is an inexpensive, yet effective tool for assessing the usability of a product, including Web sites, cell phones, interactive voice response systems, TV applications, and more. It provides an easy-to-understand score from 0 (negative) to 100 (positive). While a 100-point scale is intuitive in many respects and allows for relative judgments, information describing how the numeric score translates into an absolute judgment of usability is not known. To help answer that question, a seven-point adjective-anchored Likert scale was added as an eleventh question to nearly 1,000 SUS surveys. Results show that the Likert scale scores correlate extremely well with the SUS scores (r=0.822). The addition of the adjective rating scale to the SUS may help practitioners interpret individual SUS scores and aid in explaining the results to non-human factors professionals.
暂无摘要（点击查看原文获取完整内容）
暂无摘要（点击查看原文获取完整内容）
暂无摘要（点击查看原文获取完整内容）
Overexposure to intense sound can cause temporary or permanent hearing loss. Postexposure recovery of threshold sensitivity has been assumed to indicate reversal of damage to delicate mechano-sensory and neural structures of the inner ear and no persistent or delayed consequences for auditory function. Here, we show, using cochlear functional assays and confocal imaging of the inner ear in mouse, that acoustic overexposures causing moderate, but completely reversible, threshold elevation leave cochlear sensory cells intact, but cause acute loss of afferent nerve terminals and delayed degeneration of the cochlear nerve. Results suggest that noise-induced damage to the ear has progressive consequences that are considerably more widespread than are revealed by conventional threshold testing. This primary neurodegeneration should add to difficulties hearing in noisy environments, and could contribute to tinnitus, hyperacusis, and other perceptual anomalies commonly associated with inner ear damage.
暂无摘要（点击查看原文获取完整内容）
BACKGROUND: Vitamin K antagonists reduce the risk of stroke in patients with atrial fibrillation but are considered unsuitable in many patients, who usually receive aspirin instead. We investigated the hypothesis that the addition of clopidogrel to aspirin would reduce the risk of vascular events in patients with atrial fibrillation. METHODS: A total of 7554 patients with atrial fibrillation who had an increased risk of stroke and for whom vitamin K-antagonist therapy was unsuitable were randomly assigned to receive clopidogrel (75 mg) or placebo, once daily, in addition to aspirin. The primary outcome was the composite of stroke, myocardial infarction, non-central nervous system systemic embolism, or death from vascular causes. RESULTS: At a median of 3.6 years of follow-up, major vascular events had occurred in 832 patients receiving clopidogrel (6.8% per year) and in 924 patients receiving placebo (7.6% per year) (relative risk with clopidogrel, 0.89; 95% confidence interval [CI], 0.81 to 0.98; P=0.01). The difference was primarily due to a reduction in the rate of stroke with clopidogrel. Stroke occurred in 296 patients receiving clopidogrel (2.4% per year) and 408 patients receiving placebo (3.3% per year) (relative risk, 0.72; 95% CI, 0.62 to 0.83; P<0.001). Myocardial infarction occurred in 90 patients receiving clopidogrel (0.7% per year) and in 115 receiving placebo (0.9% per year) (relative risk, 0.78; 95% CI, 0.59 to 1.03; P=0.08). Major bleeding occurred in 251 patients receiving clopidogrel (2.0% per year) and in 162 patients receiving placebo (1.3% per year) (relative risk, 1.57; 95% CI, 1.29 to 1.92; P<0.001). CONCLUSIONS: In patients with atrial fibrillation for whom vitamin K-antagonist therapy was unsuitable, the addition of clopidogrel to aspirin reduced the risk of major vascular events, especially stroke, and increased the risk of major hemorrhage. (ClinicalTrials.gov number, NCT00249873.)
暂无摘要（点击查看原文获取完整内容）
Microtubule assembly is enhanced by the addition of 1 M sucrose or 4 M glycerol to the reassembly mixture. Tubulin can be purified from guinea pig brain readily and in good yield by two cycles of assembly in glycerol-containing solutions. The tubules assembled in glycerol and sucrose are more stable than tubules formed in the absence of these compounds. Assembly occurs in glycerol or sucrose in the absence of ATP or GTP, but is greatly accelarated by their presence.
An asymmetrically dominated alternative is dominated by one item in the set but not by another. Adding such an alternative to a choice set can increase the probability of choosing the item that dominates it. This result points to the inadequacy of many current choice models and suggests product line strategies that might not otherwise be intuitively plausible.
BACKGROUND: There is increasing interest in the potential role of the natural environment in human health and well-being. However, the evidence-base for specific and direct health or well-being benefits of activity within natural compared to more synthetic environments has not been systematically assessed. METHODS: We conducted a systematic review to collate and synthesise the findings of studies that compare measurements of health or well-being in natural and synthetic environments. Effect sizes of the differences between environments were calculated and meta-analysis used to synthesise data from studies measuring similar outcomes. RESULTS: Twenty-five studies met the review inclusion criteria. Most of these studies were crossover or controlled trials that investigated the effects of short-term exposure to each environment during a walk or run. This included 'natural' environments, such as public parks and green university campuses, and synthetic environments, such as indoor and outdoor built environments. The most common outcome measures were scores of different self-reported emotions. Based on these data, a meta-analysis provided some evidence of a positive benefit of a walk or run in a natural environment in comparison to a synthetic environment. There was also some support for greater attention after exposure to a natural environment but not after adjusting effect sizes for pretest differences. Meta-analysis of data on blood pressure and cortisol concentrations found less evidence of a consistent difference between environments across studies. CONCLUSIONS: Overall, the studies are suggestive that natural environments may have direct and positive impacts on well-being, but support the need for investment in further research on this question to understand the general significance for public health.
The structure and function of macromolecules depend critically on the ionization (protonation) states of their acidic and basic groups. A number of existing practical methods predict protonation equilibrium pK constants of macromolecules based upon their atomic resolution Protein Data Bank (PDB) structures; the calculations are often performed within the framework of the continuum electrostatics model. Unfortunately, these methodologies are complex, involve multiple steps and require considerable investment of effort. Our web server http://biophysics.cs.vt.edu/H++ provides access to a tool that automates this process, allowing both experts and novices to quickly obtain estimates of pKs as well as other related characteristics of biomolecules such as isoelectric points, titration curves and energies of protonation microstates. Protons are added to the input structure according to the calculated ionization states of its titratable groups at the user-specified pH; the output is in the PQR (PDB + charges + radii) format. In addition, corresponding coordinate and topology files are generated in the format supported by the molecular modeling package AMBER. The server is intended for a broad community of biochemists, molecular modelers, structural biologists and drug designers; it can also be used as an educational tool in biochemistry courses.