language bias

语言偏见
  • 文章类型: Journal Article
    视觉问答(VQA)任务是人工智能领域的一个重要研究方向,这需要一个可以同时理解视觉图像和自然语言问题的模型,并回答与图像有关的问题。最近的研究表明,许多视觉问答模型依赖于问题和答案之间的统计规则相关性,这反过来削弱了视觉内容和文本信息之间的相关性。在这项工作中,我们提出了一种无偏见的视觉问答方法,从加强正确答案与正面和负面预测之间的对比的角度来解决语言先验。我们设计了一个由两个具有不同角色的模块组成的新模型。我们将图像及其对应的问题输入到回答视觉注意模块中,以生成肯定的预测输出,然后使用双通道联合模块生成具有大量语言先验知识的负预测输出。最后,我们输入正面和负面的预测以及我们新设计的损失函数的正确答案进行训练。我们的方法在VQA-CPv2数据集上实现了高性能(61.24%)。此外,大多数现有的去偏置方法以降低VQA-CPv2数据集上的性能为代价,提高了VQA-CPv2数据集上的性能,而我们的方法不仅不会降低VQAv2数据集的准确性。相反,它提高了上述两个数据集上的性能。
    The Visual Question Answering (VQA) task is an important research direction in the field of artificial intelligence, which requires a model that can simultaneously understand visual images and natural language questions, and answer questions related to images. Recent studies have shown that many Visual Question Answering models rely on statistically regular correlations between questions and answers, which in turn weakens the correlation between visual content and textual information. In this work, we propose an unbiased Visual Question Answering method to solve language priors from the perspective of strengthening the contrast between the correct answer and the positive and negative predictions. We design a new model consisting of two modules with different roles. We input the image and the question corresponding to it into the Answer Visual Attention Modules to generate positive prediction output, and then use a Dual Channels Joint Module to generate negative prediction output with great linguistic prior knowledge. Finally, we input the positive and negative predictions together with the correct answer to our newly designed loss function for training. Our method achieves high performance (61.24%) on the VQA-CP v2 dataset. In addition, most existing debiasing methods improve performance on VQA-CP v2 dataset at the cost of reducing performance on VQA v2 dataset, while our method not only does not reduce the accuracy on VQA v2 dataset. Instead, it improves performance on both datasets mentioned above.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

    求助全文

  • 文章类型: Journal Article
    语言偏见是视觉问答(VQA)中值得注意的问题,其中模型往往依赖于问题和答案之间的虚假相关性来进行预测。这阻止了模型的有效推广,导致性能下降。为了解决这种偏见,我们提出了一种新的模态融合协同去偏置算法(CoD)。在我们的方法中,偏差被认为是模型在预测过程中忽略了来自特定模态的信息。我们采用协作训练方法来促进不同模态之间的相互建模,实现有效的特征融合,并使模型能够充分利用多模态知识进行预测。我们在各种数据集上的实验,包括VQA-CPv2、VQAv2和VQA-VS,使用不同的验证策略,证明我们方法的有效性。值得注意的是,采用基本基线模型对VQA-CPv2的准确率为60.14%。
    Language bias stands as a noteworthy concern in visual question answering (VQA), wherein models tend to rely on spurious correlations between questions and answers for prediction. This prevents the models from effectively generalizing, leading to a decrease in performance. In order to address this bias, we propose a novel modality fusion collaborative de-biasing algorithm (CoD). In our approach, bias is considered as the model\'s neglect of information from a particular modality during prediction. We employ a collaborative training approach to facilitate mutual modeling between different modalities, achieving efficient feature fusion and enabling the model to fully leverage multimodal knowledge for prediction. Our experiments on various datasets, including VQA-CP v2, VQA v2, and VQA-VS, using different validation strategies, demonstrate the effectiveness of our approach. Notably, employing a basic baseline model resulted in an accuracy of 60.14% on VQA-CP v2.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Pubmed)

  • 文章类型: Journal Article
    系统的评论和地图被认为是一种可靠的研究证据,但往往忽视非英语文学,这可能是重要证据的来源。要了解可能限制作者发现和包括非英语文学的能力或意图的障碍,我们评估了可能预测生态系统评价和地图中包含非英语文学的因素,以及评论作者的观点。我们评估了发表在《环境证据》(n=72)上的系统评价和地图。我们还调查了每篇论文的作者(n=32个回答),收集有关包含非英语语言文学的障碍的信息。44%的审查论文(32/72)从他们的搜索和纳入排除非英语文献。通常引用的原因包括与资源和时间有关的限制。回归分析显示,较大作者团队的评论,来自不同国家的作者,尤其是那些使用非英语主要语言的人,并且具有多语言功能的团队以更多的非英语语言进行搜索。我们的调查显示,审查团队内部的语言多样性有限,资金不足是纳入非英语语言文学的主要障碍。为了提高语言包容性并减少系统评论和地图中的偏见,我们的研究表明,评审团队内的语言多样性不断增加.将机器翻译与语言技能相结合可以减轻翻译的财务和资源负担。资助申请还可以包括翻译费用。此外,建立语言交换系统将能够访问更多语言的信息。进一步调查其他期刊中的语言收录情况的研究将加强这些结论。
    Systematic reviews and maps are considered a reliable form of research evidence, but often neglect non-English-language literature, which can be a source of important evidence. To understand the barriers that might limit authors\' ability or intent to find and include non-English-language literature, we assessed factors that may predict the inclusion of non-English-language literature in ecological systematic reviews and maps, as well as the review authors\' perspectives. We assessed systematic reviews and maps published in Environmental Evidence (n = 72). We also surveyed authors from each paper (n = 32 responses), gathering information on the barriers to the inclusion of non-English language literature. 44% of the reviewed papers (32/72) excluded non-English literature from their searches and inclusions. Commonly cited reasons included constraints related to resources and time. Regression analysis revealed that reviews with larger author teams, authors from diverse countries, especially those with non-English primary languages, and teams with multilingual capabilities searched in a significantly greater number of non-English languages. Our survey exposed limited language diversity within the review teams and inadequate funding as the principal barriers to incorporating non-English language literature. To improve language inclusion and reduce bias in systematic reviews and maps, our study suggests increasing language diversity within review teams. Combining machine translation with language skills can alleviate the financial and resource burdens of translation. Funding applications could also include translation costs. Additionally, establishing language exchange systems would enable access to information in more languages. Further studies investigating language inclusion in other journals would strengthen these conclusions.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

    求助全文

  • 文章类型: Published Erratum
    [这更正了文章DOI:10.3389/fonc.2023.1239932。].
    [This corrects the article DOI: 10.3389/fonc.2023.1239932.].
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Pubmed)

  • 文章类型: Journal Article
    全球对医疗保健的需求正在增加,在获取资源方面存在显著差异,尤其是在亚洲,非洲,和拉丁美洲。人工智能(AI)技术的快速发展,例如OpenAI的ChatGPT,在彻底改变医疗保健方面表现出了希望。然而,潜在的挑战,包括需要专门的医疗培训,隐私问题,语言偏见,需要注意。
    为了评估ChatGPT在中英文环境中的适用性和局限性,我们设计了一个实验,评估其在中国2022年国家医学许可考试(NMLE)中的表现。对于标准化评估,我们使用了NMLE的综合书面部分,由双语专家翻译成英语。所有问题都输入了ChatGPT,提供了选择它们的答案和原因。使用李克特量表评估“信息质量”的回答。
    ChatGPT显示中文的正确回答率为81.25%,英文问题的正确回答率为86.25%。Logistic回归分析表明,问题的难度和主题都不是AI错误的重要因素。Brier得分,指示预测准确性,中文为0.19,英文为0.14,表明良好的预测性能。英语回答的平均质量分数是优秀的(4.43分),略高于中国人(4.34分)。
    虽然像ChatGPT这样的AI语言模型显示了对全球医疗保健的承诺,语言偏见是一个关键挑战。确保此类技术受到严格的培训,并对多种语言和文化敏感至关重要。进一步研究AI在医疗保健中的作用,特别是在资源有限的地区,是有保证的。
    UNASSIGNED: The demand for healthcare is increasing globally, with notable disparities in access to resources, especially in Asia, Africa, and Latin America. The rapid development of Artificial Intelligence (AI) technologies, such as OpenAI\'s ChatGPT, has shown promise in revolutionizing healthcare. However, potential challenges, including the need for specialized medical training, privacy concerns, and language bias, require attention.
    UNASSIGNED: To assess the applicability and limitations of ChatGPT in Chinese and English settings, we designed an experiment evaluating its performance in the 2022 National Medical Licensing Examination (NMLE) in China. For a standardized evaluation, we used the comprehensive written part of the NMLE, translated into English by a bilingual expert. All questions were input into ChatGPT, which provided answers and reasons for choosing them. Responses were evaluated for \"information quality\" using the Likert scale.
    UNASSIGNED: ChatGPT demonstrated a correct response rate of 81.25% for Chinese and 86.25% for English questions. Logistic regression analysis showed that neither the difficulty nor the subject matter of the questions was a significant factor in AI errors. The Brier Scores, indicating predictive accuracy, were 0.19 for Chinese and 0.14 for English, indicating good predictive performance. The average quality score for English responses was excellent (4.43 point), slightly higher than for Chinese (4.34 point).
    UNASSIGNED: While AI language models like ChatGPT show promise for global healthcare, language bias is a key challenge. Ensuring that such technologies are robustly trained and sensitive to multiple languages and cultures is vital. Further research into AI\'s role in healthcare, particularly in areas with limited resources, is warranted.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Pubmed)

  • 文章类型: Journal Article
    暂无摘要。
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Pubmed)

  • 文章类型: Journal Article
    The rainfrogs of the genus Pristimantis are one of the most diverse groups of vertebrates, with outstanding reproductive modes and strategies driving their success in colonizing new habitats. The rate of Pristimantis species discovered annually has increased continuously during the last 50 years, establishing the remarkable diversity found in this genus. In this paper the specifics of publications describing new species in the group are examined, including authorship, author gender, year, language, journal, scientific collections, and other details. Detailed information on the descriptions of 591 species of Pristimantis published to date (June 2022) were analyzed and extracted. John D. Lynch and William E. Duellman are the most prolific authors, yet Latin American researchers have scaled up and continued the description processes since the 1990s. The most common language used for descriptions is English, followed by Spanish. The great majority of authors have described only one species. The largest proportion of authors who have participated in the descriptions is of Ecuadorian nationality. Ecuador is the country with the highest description rate per year (3.9% growth rate). Only 20% of the contributions have included women and only 2% have featured women as principal authors. 36.8% of the species described are in the Not Evaluated or Data Deficient categories under the IUCN global red list. The importance of enhancing the descriptions in Spanish is emphasized and the inclusion based on equal access to opportunities for female researchers in Pristimantis taxonomy is encouraged. In general, if the current trends in Pristimantis descriptions continue, in ten years, a total of 770 or more species described could be expected.
    ResumenLas ranas de la lluvia del género Pristimantis es uno de los grupos de vertebrados más diversos, con una variedad de modos reproductivos y estrategias que impulsan su éxito en la colonización de nuevos hábitats. La tasa de especies de Pristimantis descubiertas anualmente ha aumentado continuamente durante los últimos 50 años, estableciendo la notable diversidad encontrada en este género. En este artículo, examinamos los detalles de las publicaciones que describen nuevas especies en el grupo, incluida la autoría, el año, el idioma, la revista, el género, las colecciones científicas y otros detalles. Analizamos y extrajimos información detallada sobre las descripciones de 591 especies de Pristimantis publicadas hasta la fecha (junio 2022). John D. Lynch y William E. Duellman son los autores más prolíficos, pero los investigadores latinoamericanos han ampliado y continuado los procesos de descripción desde la década de 1990. El idioma más común utilizado para las descripciones es el inglés, seguido del español. La gran mayoría de los autores han descrito una sola especie. La mayor proporción de autores que han participado en las descripciones es de nacionalidad ecuatoriana. Ecuador es el país con la tasa de descripción más alta por año (tasa de crecimiento del 3,9%). Solo el 20% de las contribuciones han incluido a mujeres y solo el 2% las ha presentado como autoras principales. El 36,8% de las especies descritas se encuentran en las categorías No evaluadas o Datos insuficientes de la lista roja mundial de la UICN. Destacamos la importancia de potenciar las descripciones en español y fomentar la inclusión de mujeres investigadoras en la taxonomía de Pristimantis. En general, si continúan las tendencias actuales en las descripciones de Pristimantis, en 10 años se podría esperar un total de 770 o más especies descritas.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Pubmed)

  • 文章类型: Observational Study
    UNASSIGNED: The prevalence of inclusion of randomized controlled trials published in Latin American journals has not been evaluated yet. This study explores the extent to which randomized trials published in Latin American medical journals are cited and used in systematic reviews.
    UNASSIGNED: We did a descriptive observational study on randomized trials published in MEDLINE-indexed Latin American journals from 2010 to 2015. The primary outcome was the inclusion of these trials in systematic reviews. The secondary outcome was the total number of citations each trial received, as reported by Google Scholar.
    UNASSIGNED: Twenty-nine journals were selected. After searching these journals, we found 135 trials that fulfilled the inclusion criteria accounting for 2% of all research articles published in these journals. Of these, 55 (41%) were included in 202 systematic reviews. Of the nine most-cited randomized trials by systematic reviews and meta-analyses, only two were published in Spanish. Nine received zero citations by any article type. Most had small sample sizes.
    UNASSIGNED: The overall impact of randomized controlled trials published in Latin American journals is low. Little funding, language bias and small sample sizes may explain the low inclusion in systematic reviews and meta-analyses.
    UNASSIGNED: La prevalencia de la inclusión de ensayos controlados aleatorizados publicados en revistas latinoamericanas aún no ha sido evaluada. Este estudio tiene como objetivo explorar el grado en que los ensayos aleatorizados publicados en revistas médicas latinoamericanas son citados y utilizados en revisiones sistemáticas.
    UNASSIGNED: Se realizó un estudio observacional descriptivo sobre los ensayos aleatorizados publicados en revistas latinoamericanas indexadas en MEDLINE entre 2010 y 2015. El resultado primario fue la inclusión de estos ensayos en revisiones sistemáticas. El resultado secundario fue el número total de citas que recibió cada ensayo según lo informado por Google Scholar.
    UNASSIGNED: Se seleccionaron 29 revistas. Después de buscar en estas revistas, se encontraron 135 ensayos que cumplían los criterios de inclusión, lo que representa el 2% de todos los artículos de investigación publicados en estas revistas. De estos, 55 (41%) fueron incluidos por 202 revisiones sistemáticas. De los nueve ensayos aleatorios más citados por las revisiones sistemáticas y los metaanálisis, sólo dos fueron publicados en español. Nueve recibieron cero citas por cualquier tipo de artículo. La mayoría tenían tamaños muestrales pequeños.
    UNASSIGNED: El impacto de los ensayos controlados aleatorios publicados en revistas latinoamericanas es bajo. La escasa financiación, el sesgo lingüístico y el pequeño tamaño muestral pueden explicar la escasa inclusión en las revisiones sistemáticas y los metaanálisis.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

    求助全文

  • 文章类型: Journal Article
    背景:已经制定了系统评论(SR)中数据源的报告标准,然而,研究表明方法部分的合规性各不相同。当这种情况发生时,搜索结果的复制是困难的,并且会产生模糊和有偏见的数据源。
    目的:本研究记录了作者选择英语和非英语数据库的做法,列出所有搜索的数据库,并将学习登记处作为搜索策略的一部分。
    方法:使用分析,横截面,研究设计,志愿者数据收集器(n=107)在指定的健康状况下搜索了两个指定的英语语言平台之一的SR。在每个SR的方法部分中发现的所有数据源都被记录并使用书目技术分析模式。
    结果:审查的SR的最终样本量为N=199。在SR中看到的数据源的平均数为3.9(SD2),范围为1-10。18条记录(9%)使用单一数据源进行SR。在SRs中看到了四个领先的语言平台:英语(100%出现),高达8%使用中国数据源,4%包括西班牙语或葡萄牙语。四个最常用的数据源是:(1)Medline(98%),(2)Embase(65%),(3)科克伦图书馆(56%),和(4)WebofScience(33%)。SRs上市研究注册的百分比为30%。
    结论:减少偏见并提高SR的严谨性和可靠性的策略包括通过探索非英语语言数据库进行全面的搜索实践,使用多个数据源,搜索研究登记处。通过遵循PRISMA-S准则正确报告数据源,可以实现再现性。
    BACKGROUND: Reporting standards for data sources in systematic reviews (SRs) have been developed, yet research shows varying compliance in the methods section. When this happens, replication of search results is difficult and creates ambiguous and biased data sources.
    OBJECTIVE: This study captured author practices in choosing English and non-English-language databases, listing all the databases searched, and incorporating study registries as part of the search strategy.
    METHODS: Using an analytic, cross-sectional, study design, volunteer data collectors (n = 107) searched one of two assigned English language platforms for SRs on specified health conditions. All the data sources found in the methods section of each SR were documented and analyzed for patterns using bibliographic techniques.
    RESULTS: The final sample size of the SRs reviewed was N = 199. The mean number of data sources seen in the SRs was 3.9 (SD 2), with a range of 1-10. Eighteen records (9%) used a single data source to conduct the SRs. Four leading language platforms were seen in the SRs: English (100% of occurrences), up to 8% used Chinese data sources, and 4% included Spanish or Portuguese. The four most frequently used data sources were: (1) Medline (98%), (2) Embase (65%), (3) Cochrane Library (56%), and (4) Web of Science (33%). The percentage of SRs listing study registries was 30%.
    CONCLUSIONS: Strategies to reduce bias and increase the rigor and reliability of SRs include comprehensive search practices by exploring non-English-language databases, using multiple data sources, and searching study registries. By following PRISMA-S guidelines to report data sources correctly, reproducibility can be accomplished.
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

    求助全文

  • 文章类型: Journal Article
    暂无摘要。
    导出

    更多引用

    收藏

    翻译标题摘要

    我要上传

       PDF(Sci-hub)

公众号