Spatial understanding is a fundamental aspect of computer vision and integral for human-level reasoning about images, making it an important component for grounded language understanding. While recent large-scale text-to-image synthesis (T2I) models have shown unprecedented improvements in photorealism, it is unclear whether they have reliable spatial understanding capabilities. We investigate the ability of T2I models to generate correct spatial relationships among objects and present VISOR, an evaluation metric that captures how accurately the spatial relationship described in text is generated in the image. To benchmark existing models, we introduce a large-scale challenge dataset SR2D that contains sentences describing two objects and the spatial relationship between them. We construct and harness an automated evaluation pipeline that employs computer vision to recognize objects and their spatial relationships, and we employ it in a large-scale evaluation of T2I models. Our experiments reveal a surprising finding that, although recent state-of-the-art T2I models exhibit high image quality, they are severely limited in their ability to generate multiple objects or the specified spatial relations such as left/right/above/below. Our analyses demonstrate several biases and artifacts of T2I models such as the difficulty with generating multiple objects, a bias towards generating the first object mentioned, spatially inconsistent outputs for equivalent relationships, and a correlation between object co-occurrence and spatial understanding capabilities. We conduct a human study that shows the alignment between VISOR and human judgment about spatial understanding. We offer the SR2D dataset and the VISOR metric to the community in support of T2I spatial reasoning research.

While a substantial body of prior work has explored adversarial example generation for natural language understanding tasks, these examples are often unrealistic and diverge from the real-world data distributions. In this work, we introduce a two-stage adversarial example generation framework (NaturalAdversaries), for designing adversaries that are effective at fooling a given classifier and demonstrate natural-looking failure cases that could plausibly occur during in-the-wild deployment of the models. At the first stage a token attribution method is used to summarize a given classifier's behaviour as a function of the key tokens in the input. In the second stage a generative model is conditioned on the key tokens from the first stage. NaturalAdversaries is adaptable to both black-box and white-box adversarial attacks based on the level of access to the model parameters. Our results indicate these adversaries generalize across domains, and offer insights for future research on improving robustness of neural text classification models.

基于变压器的模型的出现，机器翻译已经快速发展。这些模型没有内置的明确的语言结构，但是它们仍然可以通过参与相关令牌隐式学习结构化的关系。我们假设通过明确赋予变形金刚具有结构性偏见，可以使这种结构学习变得更加健壮，我们研究了两种在这种偏见中构建的方法。一种方法，即TP变换器，可以增强传统的变压器体系结构，包括代表结构的附加组件。第二种方法通过将数据分割为形态令牌化来灌输数据级别的结构。我们测试了这些方法从英语翻译成土耳其语和Inuktitut的形态丰富的语言，并考虑自动指标和人类评估。我们发现，这两种方法中每种方法都允许网络实现更好的性能，但是此改进取决于数据集的大小。总而言之，结构编码方法使变压器更具样本效率，从而使它们能够从少量数据中表现得更好。

与单模式学习相比，大型数据集上的联合视觉和语言建模最近在多模式任务中表现出了良好的进步。但是，这些方法对现实世界扰动的鲁棒性尚未被研究。在这项工作中，我们对此类模型进行了首次广泛的鲁棒性研究，以针对针对视频和语言的各种现实世界的扰动。我们专注于文本到视频检索，并提出了两个大型基准数据集，即MSRVTT-P和YouCook2-P，它们利用了90个不同的视觉和35个不同的文本扰动。该研究揭示了一些有趣的发现：1）当文本受到干扰而不是视频扰动时，研究的模型更加可靠。 3）与跨注意时，使用两个分支编码器通常更健壮。我们希望这项研究能够作为基准，并指导强大的多模式学习的未来研究。

近年来，我们在视频动作识别方面取得了巨大进展。有几种基于卷积神经网络（CNN）的模型，采用了一些基于变压器的方法，可在现有基准数据集上提供最先进的性能。但是，对于这些模型，尚未研究大规模的鲁棒性，这对于现实世界应用而言是关键方面。在这项工作中，我们对这些现有模型进行大规模鲁棒性分析，以供视频识别。我们主要关注因现实世界扰动而不是对抗性扰动引起的分配变化的鲁棒性。我们提出了四个不同的基准数据集，即HMDB-51P，UCF-101P，Kinetics-400P和SSV2P，并研究了六种针对90种不同扰动的六种不同最先进的动作识别模型的鲁棒性。该研究揭示了一些有趣的发现，1）基于变压器的模型与基于CNN的模型相比，对于大多数扰动，基于变压器的模型始终更健壮，2）预训练有助于基于变压器的模型比基于CNN的模型更适合不同的扰动，而3）所有研究的模型对动力学数据集的时间扰动都具有鲁棒性，但在SSV2上却不是。这表明时间信息对于SSV2数据集的动作标签预​​测比动力学数据集更为重要。我们希望这项研究能够作为在强大的视频行动识别中进行未来研究的基准。有关该项目的更多详细信息，请访问https://rose-ar.github.io/。

有毒语言检测系统通常会错误地将包含少数群体群体提及的毒性的错误标记文本，因为这些群体通常是在线仇恨的目标。这种对虚假相关性的过度依赖也导致系统在检测隐式有毒语言方面挣扎。为了帮助缓解这些问题，我们创建了Toxigen，这是一个新的大规模和机器生成的数据集，该数据集是274K有毒和良性陈述，约有13个少数群体。我们开发了一个基于示范的提示框架和一种对抗性分类器的解码方法，以使用大量预处理的语言模型生成微妙的有毒和良性文本。以这种方式控制机器的生成使毒素可以比以前的人写文本的资源更大的规模和大约人口组覆盖隐式有毒文本。我们对毒素的一个充满挑战的子集进行人体评估，发现注释者难以区分机器生成的文本和人类写的语言。我们还发现，94.5％的有毒例子被人类注释者标记为仇恨言论。我们使用三个公开可用的数据集，我们表明，对我们的数据进行毒性分类器的填充可以大大提高其在人体编写数据上的性能。我们还证明，毒素可用于抵抗机器生成的毒性，因为鉴定在我们的评估子集中大大改善了分类器。我们的代码和数据可以在https://github.com/microsoft/toxigen上找到。

This paper presents a unified Vision-Language Pre-training (VLP) model. The model is unified in that (1) it can be finetuned for either vision-language generation (e.g., image captioning) or understanding (e.g., visual question answering) tasks, and (2) it uses a shared multi-layer transformer network for both encoding and decoding, which differs from many existing methods where the encoder and decoder are implemented using separate models. The unified VLP model is pre-trained on a large amount of image-text pairs using the unsupervised learning objectives of two tasks: bidirectional and sequence-to-sequence (seq2seq) masked vision-language prediction. The two tasks differ solely in what context the prediction conditions on. This is controlled by utilizing specific self-attention masks for the shared transformer network. To the best of our knowledge, VLP is the first reported model that achieves state-of-the-art results on both vision-language generation and understanding tasks, as disparate as image captioning and visual question answering, across three challenging benchmark datasets: COCO Captions, Flickr30k Captions, and VQA 2.0. The code and the pre-trained models are available at https://github.com/LuoweiZhou/VLP.

Research on automated essay scoring has become increasing important because it serves as a method for evaluating students' written-responses at scale. Scalable methods for scoring written responses are needed as students migrate to online learning environments resulting in the need to evaluate large numbers of written-response assessments. The purpose of this study is to describe and evaluate three active learning methods than can be used to minimize the number of essays that must be scored by human raters while still providing the data needed to train a modern automated essay scoring system. The three active learning methods are the uncertainty-based, the topological-based, and the hybrid method. These three methods were used to select essays included as part of the Automated Student Assessment Prize competition that were then classified using a scoring model that was training with the bidirectional encoder representations from transformer language model. All three active learning methods produced strong results, with the topological-based method producing the most efficient classification. Growth rate accuracy was also evaluated. The active learning methods produced different levels of efficiency under different sample size allocations but, overall, all three methods were highly efficient and produced classifications that were similar to one another.

Osteoarthritis (OA) is the most prevalent chronic joint disease worldwide, where knee OA takes more than 80% of commonly affected joints. Knee OA is not a curable disease yet, and it affects large columns of patients, making it costly to patients and healthcare systems. Etiology, diagnosis, and treatment of knee OA might be argued by variability in its clinical and physical manifestations. Although knee OA carries a list of well-known terminology aiming to standardize the nomenclature of the diagnosis, prognosis, treatment, and clinical outcomes of the chronic joint disease, in practice there is a wide range of terminology associated with knee OA across different data sources, including but not limited to biomedical literature, clinical notes, healthcare literacy, and health-related social media. Among these data sources, the scientific articles published in the biomedical literature usually make a principled pipeline to study disease. Rapid yet, accurate text mining on large-scale scientific literature may discover novel knowledge and terminology to better understand knee OA and to improve the quality of knee OA diagnosis, prevention, and treatment. The present works aim to utilize artificial neural network strategies to automatically extract vocabularies associated with knee OA diseases. Our finding indicates the feasibility of developing word embedding neural networks for autonomous keyword extraction and abstraction of knee OA.

Vision transformers (ViTs) are quickly becoming the de-facto architecture for computer vision, yet we understand very little about why they work and what they learn. While existing studies visually analyze the mechanisms of convolutional neural networks, an analogous exploration of ViTs remains challenging. In this paper, we first address the obstacles to performing visualizations on ViTs. Assisted by these solutions, we observe that neurons in ViTs trained with language model supervision (e.g., CLIP) are activated by semantic concepts rather than visual features. We also explore the underlying differences between ViTs and CNNs, and we find that transformers detect image background features, just like their convolutional counterparts, but their predictions depend far less on high-frequency information. On the other hand, both architecture types behave similarly in the way features progress from abstract patterns in early layers to concrete objects in late layers. In addition, we show that ViTs maintain spatial information in all layers except the final layer. In contrast to previous works, we show that the last layer most likely discards the spatial information and behaves as a learned global pooling operation. Finally, we conduct large-scale visualizations on a wide range of ViT variants, including DeiT, CoaT, ConViT, PiT, Swin, and Twin, to validate the effectiveness of our method.