大型语言模型（LLM）已在一系列自然语言理解任务上实现了最先进的表现。但是，这些LLM可能依靠数据集偏差和文物作为预测的快捷方式。这极大地损害了他们的分布（OOD）概括和对抗性鲁棒性。在本文中，我们对最新发展的综述，这些发展解决了LLMS的鲁棒性挑战。我们首先介绍LLM的概念和鲁棒性挑战。然后，我们介绍了在LLM中识别快捷方式学习行为的方法，表征了快捷方式学习的原因以及引入缓解解决方案。最后，我们确定了关键挑战，并将这一研究线的联系引入其他方向。

由于NLP模型实现了基准测试的最先进的性能并获得了广泛的应用程序，因此确保在现实世界中的安全部署这些模型的安全部署，例如，确保模型对未经调用或具有挑战性的情景稳健。尽管具有越来越多的学习主题，但它在视觉和NLP等应用中分别探讨了，具有多种研究中的各种定义，评估和缓解策略。在本文中，我们的目标是提供对如何定义，测量和提高NLP鲁棒性的统一调查。我们首先连接多种稳健性的定义，然后统一各种各样的工作方面识别稳健性失败和评估模型的鲁棒性。相应地，我们呈现了数据驱动，模型驱动和基于归纳的缓解策略，具有如何有效地改善NLP模型中的鲁棒性的更系统的观点。最后，我们通过概述开放的挑战和未来方向来促进在这一领域的进一步研究。

最近的自然语言处理（NLP）技术在基准数据集中实现了高性能，主要原因是由于深度学习性能的显着改善。研究界的进步导致了最先进的NLP任务的生产系统的巨大增强，例如虚拟助理，语音识别和情感分析。然而，随着对抗性攻击测试时，这种NLP系统仍然仍然失败。初始缺乏稳健性暴露于当前模型的语言理解能力中的令人不安的差距，当NLP系统部署在现实生活中时，会产生问题。在本文中，我们通过以各种维度的系统方式概述文献来展示了NLP稳健性研究的结构化概述。然后，我们深入了解稳健性的各种维度，跨技术，指标，嵌入和基准。最后，我们认为，鲁棒性应该是多维的，提供对当前研究的见解，确定文学中的差距，以建议值得追求这些差距的方向。

快捷方式学习的问题在NLP中广为人知，并且近年来一直是重要的研究重点。数据中的意外相关性使模型能够轻松地求解旨在表现出高级语言理解和推理能力的任务。在本调查论文中，我们关注机器阅读理解的领域（MRC），这是展示高级语言理解的重要任务，这也遭受了一系列快捷方式。我们总结了用于测量和减轻快捷方式的可用技术，并以捷径研究进一步进展的建议结论。最重要的是，我们强调了MRC中缓解快捷方式的两个主要问题：缺乏公共挑战集，有效和可重复使用的评估的必要组成部分以及在其他领域中缺乏某些缓解技术。

具有大量偏见的数据集当前威胁要培训有关NLU任务的值得信赖的模型。尽管取得了巨大进展，但当前的偏见方法却过分依赖偏见属性的知识。但是，属性的​​定义是难以捉摸的，并且在不同的数据集上有所不同。此外，利用输入级别的这些属性到偏置缓解可能会留下内在属性与基本决策规则之间的差距。为了缩小这一差距并解放有关偏见的监督，我们建议将缓解偏见扩展到特征空间。因此，开发了一个新型模型，即恢复具有无知识（风险）的预期功能子空间。假设由各种偏见引起的快捷键特征是为了预测而无意的，则风险将其视为冗余特征。当研究较低的歧管以去除冗余时，风险表明，具有预期功能的极低维度子空间可以牢固地表示高度偏见的数据集。经验结果表明，我们的模型可以始终如一地提高模型的概括到分布式集合，并实现新的最新性能。

微调预训练模型在标准的自然语言处理基准上取得了令人印象深刻的性能。然而，所产生的模型概括性仍然明确地理解。例如，我们不知道，性能如何导致泛化模型的完善。在这项研究中，我们使用关系提取来分析来自不同观点的微调BERT模型。我们还根据我们提出的改进来表征泛化技术的差异。从经验实验中，我们发现BERT通过随机化，对抗性和反事实试验以及偏差（即选择和语义）遭受鲁棒性而遭受瓶颈。这些发现突出了未来改进的机会。我们的开放式测试平台诊断为\ url {https://github.com/zjunlp/diagnosere}。

神经网络通常使预测依赖于数据集的虚假相关性，而不是感兴趣的任务的内在特性，面对分布外（OOD）测试数据的急剧下降。现有的De-Bias学习框架尝试通过偏置注释捕获特定的DataSet偏差，它们无法处理复杂的“ood方案”。其他人在低能力偏置模型或损失上隐含地识别数据集偏置，但在训练和测试数据来自相同分布时，它们会降低。在本文中，我们提出了一般的贪婪去偏见学习框架（GGD），它贪婪地训练偏置模型和基础模型，如功能空间中的梯度下降。它鼓励基础模型专注于用偏置模型难以解决的示例，从而仍然在测试阶段中的杂散相关性稳健。 GGD在很大程度上提高了各种任务的模型的泛化能力，但有时会过度估计偏置水平并降低在分配测试。我们进一步重新分析了GGD的集合过程，并将课程正规化为由课程学习启发的GGD，这取得了良好的分配和分发性能之间的权衡。对图像分类的广泛实验，对抗问题应答和视觉问题应答展示了我们方法的有效性。 GGD可以在特定于特定于任务的偏置模型的设置下学习更强大的基础模型，其中具有现有知识和自组合偏置模型而无需先验知识。

It has been shown that NLI models are usually biased with respect to the word-overlap between premise and hypothesis; they take this feature as a primary cue for predicting the entailment label. In this paper, we focus on an overlooked aspect of the overlap bias in NLI models: the reverse word-overlap bias. Our experimental results demonstrate that current NLI models are highly biased towards the non-entailment label on instances with low overlap, and the existing debiasing methods, which are reportedly successful on existing challenge datasets, are generally ineffective in addressing this category of bias. We investigate the reasons for the emergence of the overlap bias and the role of minority examples in its mitigation. For the former, we find that the word-overlap bias does not stem from pre-training, and for the latter, we observe that in contrast to the accepted assumption, eliminating minority examples does not affect the generalizability of debiasing methods with respect to the overlap bias.

大规模的预训练语言模型在广泛的自然语言理解（NLU）任务中取得了巨大的成功，甚至超过人类性能。然而，最近的研究表明，这些模型的稳健性可能受到精心制作的文本对抗例子的挑战。虽然已经提出了几个单独的数据集来评估模型稳健性，但仍缺少原则和全面的基准。在本文中，我们呈现对抗性胶水（AdvGlue），这是一个新的多任务基准，以定量和彻底探索和评估各种对抗攻击下现代大规模语言模型的脆弱性。特别是，我们系统地应用14种文本对抗的攻击方法来构建一个粘合的援助，这是由人类进一步验证的可靠注释。我们的调查结果总结如下。 （i）大多数现有的对抗性攻击算法容易发生无效或暧昧的对手示例，其中大约90％的含量改变原始语义含义或误导性的人的注册人。因此，我们执行仔细的过滤过程来策划高质量的基准。 （ii）我们测试的所有语言模型和强大的培训方法在AdvGlue上表现不佳，差价远远落后于良性准确性。我们希望我们的工作能够激励开发新的对抗攻击，这些攻击更加隐身，更加统一，以及针对复杂的对抗性攻击的新强大语言模型。 Advglue在https://adversarialglue.github.io提供。

Graph machine learning has been extensively studied in both academia and industry. Although booming with a vast number of emerging methods and techniques, most of the literature is built on the in-distribution hypothesis, i.e., testing and training graph data are identically distributed. However, this in-distribution hypothesis can hardly be satisfied in many real-world graph scenarios where the model performance substantially degrades when there exist distribution shifts between testing and training graph data. To solve this critical problem, out-of-distribution (OOD) generalization on graphs, which goes beyond the in-distribution hypothesis, has made great progress and attracted ever-increasing attention from the research community. In this paper, we comprehensively survey OOD generalization on graphs and present a detailed review of recent advances in this area. First, we provide a formal problem definition of OOD generalization on graphs. Second, we categorize existing methods into three classes from conceptually different perspectives, i.e., data, model, and learning strategy, based on their positions in the graph machine learning pipeline, followed by detailed discussions for each category. We also review the theories related to OOD generalization on graphs and introduce the commonly used graph datasets for thorough evaluations. Finally, we share our insights on future research directions. This paper is the first systematic and comprehensive review of OOD generalization on graphs, to the best of our knowledge.

Despite being responsible for state-of-the-art results in several computer vision and natural language processing tasks, neural networks have faced harsh criticism due to some of their current shortcomings. One of them is that neural networks are correlation machines prone to model biases within the data instead of focusing on actual useful causal relationships. This problem is particularly serious in application domains affected by aspects such as race, gender, and age. To prevent models from incurring on unfair decision-making, the AI community has concentrated efforts in correcting algorithmic biases, giving rise to the research area now widely known as fairness in AI. In this survey paper, we provide an in-depth overview of the main debiasing methods for fairness-aware neural networks in the context of vision and language research. We propose a novel taxonomy to better organize the literature on debiasing methods for fairness, and we discuss the current challenges, trends, and important future work directions for the interested researcher and practitioner.

众包NLP数据集的反复挑战是，在制作示例时，人类作家通常会依靠重复的模式，从而导致缺乏语言多样性。我们介绍了一种基于工人和AI协作的数据集创建的新方法，该方法汇集了语言模型的生成力量和人类的评估力量。从现有的数据集，自然语言推理（NLI）的Multinli开始，我们的方法使用数据集制图自动识别示例来证明具有挑战性的推理模式，并指示GPT-3撰写具有相似模式的新示例。然后，机器生成的示例会自动过滤，并最终由人类人群工人修订和标记。最终的数据集Wanli由107,885个NLI示例组成，并在现有的NLI数据集上呈现出独特的经验优势。值得注意的是，培训有关Wanli的模型，而不是Multinli（$ 4 $ $倍）可改善我们考虑的七个外域测试集的性能，包括汉斯（Hans）的11％和对抗性NLI的9％。此外，将Multinli与Wanli结合起来比将其与其他NLI增强集相结合更有效。我们的结果表明，自然语言生成技术的潜力是策划增强质量和多样性的NLP数据集。

Transformer-based models have pushed state of the art in many areas of NLP, but our understanding of what is behind their success is still limited. This paper is the first survey of over 150 studies of the popular BERT model. We review the current state of knowledge about how BERT works, what kind of information it learns and how it is represented, common modifications to its training objectives and architecture, the overparameterization issue and approaches to compression. We then outline directions for future research.

可解释的人工智能（XAI）中方法的动机通常包括检测，量化和缓解偏见，并为使机器学习模型更加公平而做出贡献。但是，确切的XAI方法可以如何帮助打击偏见。在本文中，我们简要回顾了NLP研究中的解释性和公平性的趋势，确定了当前的实践，其中采用了解释性方法来检测和减轻偏见，并调查了阻止XAI方法在解决公平问题中更广泛使用的障碍。

As the societal impact of Deep Neural Networks (DNNs) grows, the goals for advancing DNNs become more complex and diverse, ranging from improving a conventional model accuracy metric to infusing advanced human virtues such as fairness, accountability, transparency (FaccT), and unbiasedness. Recently, techniques in Explainable Artificial Intelligence (XAI) are attracting considerable attention, and have tremendously helped Machine Learning (ML) engineers in understanding AI models. However, at the same time, we started to witness the emerging need beyond XAI among AI communities; based on the insights learned from XAI, how can we better empower ML engineers in steering their DNNs so that the model's reasonableness and performance can be improved as intended? This article provides a timely and extensive literature overview of the field Explanation-Guided Learning (EGL), a domain of techniques that steer the DNNs' reasoning process by adding regularization, supervision, or intervention on model explanations. In doing so, we first provide a formal definition of EGL and its general learning paradigm. Secondly, an overview of the key factors for EGL evaluation, as well as summarization and categorization of existing evaluation procedures and metrics for EGL are provided. Finally, the current and potential future application areas and directions of EGL are discussed, and an extensive experimental study is presented aiming at providing comprehensive comparative studies among existing EGL models in various popular application domains, such as Computer Vision (CV) and Natural Language Processing (NLP) domains.

Models trained via empirical risk minimization (ERM) are known to rely on spurious correlations between labels and task-independent input features, resulting in poor generalization to distributional shifts. Group distributionally robust optimization (G-DRO) can alleviate this problem by minimizing the worst-case loss over a set of pre-defined groups over training data. G-DRO successfully improves performance of the worst-group, where the correlation does not hold. However, G-DRO assumes that the spurious correlations and associated worst groups are known in advance, making it challenging to apply it to new tasks with potentially multiple unknown spurious correlations. We propose AGRO -- Adversarial Group discovery for Distributionally Robust Optimization -- an end-to-end approach that jointly identifies error-prone groups and improves accuracy on them. AGRO equips G-DRO with an adversarial slicing model to find a group assignment for training examples which maximizes worst-case loss over the discovered groups. On the WILDS benchmark, AGRO results in 8% higher model performance on average on known worst-groups, compared to prior group discovery approaches used with G-DRO. AGRO also improves out-of-distribution performance on SST2, QQP, and MS-COCO -- datasets where potential spurious correlations are as yet uncharacterized. Human evaluation of ARGO groups shows that they contain well-defined, yet previously unstudied spurious correlations that lead to model errors.

众所周知，端到端的神经NLP体系结构很难理解，这引起了近年来为解释性建模的许多努力。模型解释的基本原则是忠诚，即，解释应准确地代表模型预测背后的推理过程。这项调查首先讨论了忠诚的定义和评估及其对解释性的意义。然后，我们通过将方法分为五类来介绍忠实解释的最新进展：相似性方法，模型内部结构的分析，基于反向传播的方法，反事实干预和自我解释模型。每个类别将通过其代表性研究，优势和缺点来说明。最后，我们从它们的共同美德和局限性方面讨论了上述所有方法，并反思未来的工作方向忠实的解释性。对于有兴趣研究可解释性的研究人员，这项调查将为该领域提供可访问且全面的概述，为进一步探索提供基础。对于希望更好地了解自己的模型的用户，该调查将是一项介绍性手册，帮助选择最合适的解释方法。

Recent methods demonstrate that data augmentation using counterfactual knowledge can teach models the causal structure of a task, leading to robust and generalizable models. However, such counterfactual data often has a limited scale and diversity if crowdsourced and is computationally expensive to extend to new perturbation types if generated using supervised methods. To address this, we introduce a new framework called DISCO for automatically generating high-quality counterfactual data at scale. DISCO engineers prompts to generate phrasal perturbations with a large general language model. Then, a task-specific teacher model filters the generation to distill high-quality counterfactual data. We show that learning with this counterfactual data yields a comparatively small student model that is 6% (absolute) more robust and generalizes 5% better across distributions than baselines on various challenging evaluations. This model is also 15% more sensitive in differentiating original and counterfactual examples, on three evaluation sets written by human workers and via human-AI collaboration.

Machine learning models rely on various assumptions to attain high accuracy. One of the preliminary assumptions of these models is the independent and identical distribution, which suggests that the train and test data are sampled from the same distribution. However, this assumption seldom holds in the real world due to distribution shifts. As a result models that rely on this assumption exhibit poor generalization capabilities. Over the recent years, dedicated efforts have been made to improve the generalization capabilities of these models collectively known as -- \textit{domain generalization methods}. The primary idea behind these methods is to identify stable features or mechanisms that remain invariant across the different distributions. Many generalization approaches employ causal theories to describe invariance since causality and invariance are inextricably intertwined. However, current surveys deal with the causality-aware domain generalization methods on a very high-level. Furthermore, we argue that it is possible to categorize the methods based on how causality is leveraged in that method and in which part of the model pipeline is it used. To this end, we categorize the causal domain generalization methods into three categories, namely, (i) Invariance via Causal Data Augmentation methods which are applied during the data pre-processing stage, (ii) Invariance via Causal representation learning methods that are utilized during the representation learning stage, and (iii) Invariance via Transferring Causal mechanisms methods that are applied during the classification stage of the pipeline. Furthermore, this survey includes in-depth insights into benchmark datasets and code repositories for domain generalization methods. We conclude the survey with insights and discussions on future directions.

Question answering (QA) models for reading comprehension tend to learn shortcut solutions rather than the solutions intended by QA datasets. QA models that have learned shortcut solutions can achieve human-level performance in shortcut examples where shortcuts are valid, but these same behaviors degrade generalization potential on anti-shortcut examples where shortcuts are invalid. Various methods have been proposed to mitigate this problem, but they do not fully take the characteristics of shortcuts themselves into account. We assume that the learnability of shortcuts, i.e., how easy it is to learn a shortcut, is useful to mitigate the problem. Thus, we first examine the learnability of the representative shortcuts on extractive and multiple-choice QA datasets. Behavioral tests using biased training sets reveal that shortcuts that exploit answer positions and word-label correlations are preferentially learned for extractive and multiple-choice QA, respectively. We find that the more learnable a shortcut is, the flatter and deeper the loss landscape is around the shortcut solution in the parameter space. We also find that the availability of the preferred shortcuts tends to make the task easier to perform from an information-theoretic viewpoint. Lastly, we experimentally show that the learnability of shortcuts can be utilized to construct an effective QA training set; the more learnable a shortcut is, the smaller the proportion of anti-shortcut examples required to achieve comparable performance on shortcut and anti-shortcut examples. We claim that the learnability of shortcuts should be considered when designing mitigation methods.