In this work we ask whether it is possible to create a "universal" detector for telling apart real images from these generated by a CNN, regardless of architecture or dataset used. To test this, we collect a dataset consisting of fake images generated by 11 different CNN-based image generator models, chosen to span the space of commonly used architectures today (ProGAN, StyleGAN, BigGAN, CycleGAN, StarGAN, GauGAN, DeepFakes, cascaded refinement networks, implicit maximum likelihood estimation, second-order attention super-resolution, seeing-in-the-dark). We demonstrate that, with careful pre- and post-processing and data augmentation, a standard image classifier trained on only one specific CNN generator (ProGAN) is able to generalize surprisingly well to unseen architectures, datasets, and training methods (including the just released StyleGAN2). Our findings suggest the intriguing possibility that today's CNN-generated images share some common systematic flaws, preventing them from achieving realistic image synthesis.
ADVERTISEMENT RETURN TO ISSUEPREVArticleNEXTBis(imino)pyridines: Surprisingly Reactive Ligands and a Gateway to New Families of CatalystsVernon C. Gibson, Carl Redshaw, and Gregory A. SolanView Author Information Department of Chemistry, Imperial College, South Kensington Campus, London, SW7 2AZ, United Kingdom, Wolfson Materials and Catalysis Centre, School of Chemical Sciences and Pharmacy, The University of East Anglia, Norwich, NR4 7TJ, United Kingdom, and Department of Chemistry, University of Leicester, University Road, Leicester, LE1 7RH, United Kingdom Cite this: Chem. Rev. 2007, 107, 5, 1745–1776Publication Date (Web):May 9, 2007Publication History Received12 October 2006Published online9 May 2007Published inissue 1 May 2007https://doi.org/10.1021/cr068437yCopyright © 2007 American Chemical SocietyRIGHTS & PERMISSIONSArticle Views8748Altmetric-Citations726LEARN ABOUT THESE METRICSArticle Views are the COUNTER-compliant sum of full text article downloads since November 2008 (both PDF and HTML) across all institutions and individuals. These metrics are regularly updated to reflect usage leading up to the last few days.Citations are the number of other articles citing this article, calculated by Crossref and updated daily. Find more information about Crossref citation counts.The Altmetric Attention Score is a quantitative measure of the attention that a research article has received online. Clicking on the donut icon will load a page at altmetric.com with additional details about the score and the social media presence for the given article. Find more information on the Altmetric Attention Score and how the score is calculated. Share Add toView InAdd Full Text with ReferenceAdd Description ExportRISCitationCitation and abstractCitation and referencesMore Options Share onFacebookTwitterWechatLinked InReddit Read OnlinePDF (880 KB) Get e-AlertsSUBJECTS:Alkyls,Anions,Catalysts,Iron,Ligands Get e-Alerts
We measured the expression pattern and analyzed codon usage in 8,133, 1,550, and 2,917 genes, respectively, from Caenorhabditis elegans, Drosophila melanogaster, and Arabidopsis thaliana. In those three species, we observed a clear correlation between codon usage and gene expression levels and showed that this correlation is not due to a mutational bias. This provides direct evidence for selection on silent sites in those three distantly related multicellular eukaryotes. Surprisingly, there is a strong negative correlation between codon usage and protein length. This effect is not due to a smaller size of highly expressed proteins. Thus, for a same-expression pattern, the selective pressure on codon usage appears to be lower in genes encoding long rather than short proteins. This puzzling observation is not predicted by any of the current models of selection on codon usage and thus raises the question of how translation efficiency affects fitness in multicellular organisms.
This paper argues that using a set of semistructured message templates is surprisingly helpful in designing a variety of computer-based communication and coordination systems. Semistructured messages can help provide automatic aids for (1) composing messages to be sent, (2) selecting, sorting, and prioritizing messages that are received, (3) responding automatically to some messages, and (4) suggesting likely responses to other messages. The use of these capabilities is illustrated in a range of applications including electronic mail, computer conferencing, calendar management, and task tracking. The applications show how ideas from artificial intelligence (such as inheritance and production rules) and ideas from user interface design (such as interactive graphical editors) can be combined in novel ways for dealing with semistructured messages. The final part of the paper discusses how communities can evolve a useful set of message type definitions.
It is widely believed that to efficiently represent an otherwise smooth object with discontinuities along edges, one must use an adaptive representation that in some sense `tracks' the shape of the discontinuity set. This folk-belief -- some would say folk-theorem -- is incorrect. At the very least, the possible quantitative advantage of such adaptation is vastly smaller than commonly believed. We have recently constructed a tight frame of curvelets which provides stable, efficient, and near-optimal representation of otherwise smooth objects having discontinuities along smooth curves. By applying naive thresholding to the curvelet transform of such an object, one can form m-term approximations with rate of L² approximation rivaling the rate obtainable by complex adaptive schemes which attempt to `track' the discontinuity set. In this article we explain the basic issues of efficient m-term approximation, the construction of efficient adaptive representation, the construction ...
This paper links the sharp drop in US manufacturing employment after 2000 to a change in US trade policy that eliminated potential tariff increases on Chinese imports. Industries more exposed to the change experience greater employment loss, increased imports from China, and higher entry by US importers and foreign-owned Chinese exporters. At the plant level, shifts toward less labor-intensive production and exposure to the policy via input-output linkages also contribute to the decline in employment. Results are robust to other potential explanations of employment loss, and there is no similar reaction in the European Union, where policy did not change. (JEL D72, E24, F13, F16, L24, L60, P33)
We report an efficient synthesis of copper indium sulfide nanocrystals with strong photoluminescence in the visible to near-infrared. This method can produce gram quantities of material with a chemical yield in excess of 90% with minimal solvent waste. The overgrowth of as-prepared nanocrystals with a few monolayers of CdS or ZnS increases the photoluminescence quantum efficiency to > 80%. On the basis of time-resolved spectroscopic studies of core/shell particles, we conclude that the emission is due to an optical transition that couples a quantized electron state to a localized hole state, which is most likely associated with an internal defect.
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）
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）
Academics, politicians, and journalists are often highly critical of U.S . firms for holding too much cash. Cash holdings are stockpiled free‐cash flow and incur substantial opportunity costs from the perspectives of economics. However, behavioral theory highlights the benefits of cash holdings as fungible slack resources facilitating adaptive advantages. We use the countervailing forces embodied in these two theories to hypothesize and test a quadratic functional relationship of returns to cash measured by T obin's q. We also build and test a related novel hypothesis of scale‐dependent returns to cash based on the competitive strategy concept of strategic deterrence. Tests for both of these hypotheses are positive and show that returns to cash continue to increase far beyond transactional needs . Copyright © 2013 John Wiley & Sons, Ltd.
Proximal Policy Optimization (PPO) is a ubiquitous on-policy reinforcement learning algorithm but is significantly less utilized than off-policy learning algorithms in multi-agent settings. This is often due to the belief that PPO is significantly less sample efficient than off-policy methods in multi-agent systems. In this work, we carefully study the performance of PPO in cooperative multi-agent settings. We show that PPO-based multi-agent algorithms achieve surprisingly strong performance in four popular multi-agent testbeds: the particle-world environments, the StarCraft multi-agent challenge, Google Research Football, and the Hanabi challenge, with minimal hyperparameter tuning and without any domain-specific algorithmic modifications or architectures. Importantly, compared to competitive off-policy methods, PPO often achieves competitive or superior results in both final returns and sample efficiency. Finally, through ablation studies, we analyze implementation and hyperparameter factors that are critical to PPO's empirical performance, and give concrete practical suggestions regarding these factors. Our results show that when using these practices, simple PPO-based methods can be a strong baseline in cooperative multi-agent reinforcement learning. Source code is released at \url{https://github.com/marlbenchmark/on-policy}.
Metrics for evaluating generative models aim to measure the discrepancy between real and generated images. The often-used Fréchet Inception Distance (FID) metric, for example, extracts “high-level” features using a deep network from the two sets. However, we find that the differences in “low-level” preprocessing, specifically image resizing and compression, can induce large variations and have unforeseen consequences. For instance, when resizing an image, e.g., with a bilinear or bicubic kernel, signal processing principles mandate adjusting prefilter width depending on the downsampling factor, to antialias to the appropriate bandwidth. However, commonly-used implementations use a fixed-width prefilter, resulting in aliasing artifacts. Such aliasing leads to corruptions in the feature extraction down-stream. Next, lossy compression, such as JPEG, is commonly used to reduce the file size of an image. Although designed to minimally degrade the perceptual quality of an image, the operation also produces variations downstream. Furthermore, we show that if compression is used on real training images, FID can actually improve if the generated images are also subsequently compressed. This paper shows that choices in low-level image processing have been an under-appreciated aspect of generative modeling. We identify and characterize variations in generative modeling development pipelines, provide recommendations based on signal processing principles, and release a reference implementation to facilitate future comparisons.
"As a cohort of people, animals, or machines ages, the individuals at highest risk tend to die or exit first. This differential selection can produce patterns of mortality for the population as a whole that are surprisingly different from the patterns for subpopulations or individuals. Naive acceptance of observed population patterns may lead to erroneous policy recommendations if an intervention depends on the response of individuals. Furthermore, because patterns at the individual level may be simpler than composite population patterns, both theoretical and empirical research may be unnecessarily complicated by failure to recognize the effects of heterogeneity."
A measure of preference for consistency (the PFC Scale) was developed. In three construct validation experiments, scores on the PFC successfully predicted individuals who would and would not be susceptible to a set of standard consistency-based effects: cognitive balance, foot in the door, and dissonance. The pattern of results in each of the experiments suggested the type of consistency that the PFC measures: a tendency to base one's responses to incoming stimuli on the implications of existing (prior entry) variables, such as previous expectancies, commitments, and choices. A surprisingly large percentage (at least half) of our participants showed no strong inherent preference for consistency—a finding that may explain the frequent failure to detect or replicate (a) traditional consistency effects and (b) a wide variety of other experimental phenomena. It is quite remarkable how many of the great, early theorists of social psychology developed or played an important role in developing consistency theories of human motivation (Brehm& Cohen, 1962;Festinger, 1957;Heider, 1946, 1958; McGuire, 1960; Newcomb, 1953; Osgood & Tannenbaum, 1955; Rokeach, 1960; Rosenberg & Abelson, 1960; Zajonc, 1960). It seems equally remarkable, then, that despite the counsel of so respected an array of minds and despite a decade-long cornucopia of supportive work (from the late 1950s to the late 1960s), consistency-based explanations are rarely invoked in present-day social psychological accounts of human functioning (Aronson, 1992; Berkowitz & Devine, 1989), even though several of the faithful have continued through the years to try to draw attention to the applicability of consistency motives to a wide array of behavior (Aronson, 1992; Cialdini & DeNicholas, 1989; McGuire, 1990; Wicklund & Brehm, 1976;Insko, 1984). To what are we to attribute this rather dramatic shift by social psychologists away from the explanatory utility of the consistency principle—that people are motivated toward cognitive consistency and will change their beliefs, attitudes, perceptions, and actions to achieve it? On a general level, one could point to the often decried trendiness of scientific investigation and to the reward structure of the scientific enterprise, which make well-researched issues less
Secondary analyses of Revised NEO Personality Inventory data from 26 cultures (N = 23,031) suggest that gender differences are small relative to individual variation within genders; differences are replicated across cultures for both college-age and adult samples, and differences are broadly consistent with gender stereotypes: Women reported themselves to be higher in Neuroticism, Agreeableness, Warmth, and Openness to Feelings, whereas men were higher in Assertiveness and Openness to Ideas. Contrary to predictions from evolutionary theory, the magnitude of gender differences varied across cultures. Contrary to predictions from the social role model, gender differences were most pronounced in European and American cultures in which traditional sex roles are minimized. Possible explanations for this surprising finding are discussed, including the attribution of masculine and feminine behaviors to roles rather than traits in traditional cultures.
暂无摘要（点击查看原文获取完整内容）
Assumptions regarding the importance of empathy are pervasive. Given the impact these assumptions have on research, assessment, and treatment, it is imperative to know whether they are valid. Of particular interest is a basic question: Are deficits in empathy associated with aggressive behavior? Previous attempts to review the relation between empathy and aggression yielded inconsistent results and generally included a small number of studies. To clarify these divergent findings, we comprehensively reviewed the relation of empathy to aggression in adults, including community, student, and criminal samples. A mixed effects meta-analysis of published and unpublished studies involving 106 effect sizes revealed that the relation between empathy and aggression was surprisingly weak (r = -.11). This finding was fairly consistent across specific types of aggression, including verbal aggression (r = -.20), physical aggression (r = -.12), and sexual aggression (r = -.09). Several potentially important moderators were examined, although they had little impact on the total effect size. The results of this study are particularly surprising given that empathy is a core component of many treatments for aggressive offenders and that most psychological disorders of aggression include diagnostic criteria specific to deficient empathic responding. We discuss broad conclusions, consider implications for theory, and address current limitations in the field, such as reliance on a small number of self-report measures of empathy. We highlight the need for diversity in measurement and suggest a new operationalization of empathy that may allow it to synchronize with contemporary thinking regarding its role in aggressive behavior.