近年来，深入的强化学习（RL）在各种组合搜索领域（例如两人游戏和科学发现）中都取得了成功。但是，直接在计划域中应用深度RL仍然具有挑战性。一个主要的困难是，如果没有人工制作的启发式功能，奖励信号除非学习框架发现任何解决方案计划，否则奖励信号将保持零。随着计划的最小长度的增长，搜索空间变为\ emph {指数更大}，这是计划实例的严重限制，该实例的计划最小计划长度为数百到数千步。以前的学习框架可以增强使用深神经网络和额外生成的子观念的图形搜索在各种具有挑战性的计划域中取得了成功。但是，生成有用的子目标需要广泛的领域知识。我们提出了一种独立于域的方法，该方法可以通过图值迭代来增强图形搜索，以求解针对域特有的求解器无法实现的硬计划实例。特别是，我们的方法还没有仅从发现的计划中获得学习信号，而是从未达到目标状态的失败尝试中学习。图值迭代组件可以利用本地搜索空间的图形结构并提供更有信息的学习信号。我们还展示了如何使用课程策略来平滑学习过程并对图形值迭代量表的完整分析并实现学习。

尽管实用的求解器在各种NP完整域中取得了成功，例如SAT和CSP以及使用深度强化学习来解决诸如GO之类的两人游戏，但某些类别的Pspace-Hard计划问题仍然遥不可及。由于硬实例的指数搜索空间，即使是精心设计的域专用求解器也可能会迅速失败。结合了传统搜索方法的最新作品，例如最佳优先搜索和蒙特卡洛树搜索，以及深度神经网络（DNN）的启发式方法，已经显示出有希望的进步，并且可以解决超出专业求解器以外的大量艰苦计划实例。为了更好地理解这些方法为何起作用，我们研究了基于DNN的最佳优先搜索的政策和价值网络的相互作用，并展示了该策略网络的惊人有效性，并通过价值网络进一步增强了价值网络，作为指导启发式的启发式启发式程序。搜索。为了进一步理解现象，我们研究了搜索算法的成本分布，发现索科巴实例可以具有重尾的运行时分布，左侧和右侧都有尾巴。特别是，我们首次展示了\ textit {左尾巴}的存在，并提出了一个抽象的树模型，可以从经验上解释这些尾巴的外观。该实验表明，政策网络是一种强大的启发式指导搜索的关键作用，这可以通过避免探索成倍的尺寸的子树来导致左尾部具有多项式缩放。我们的结果还证明了与传统组合求解器中广泛使用的随机重新启动的重要性，用于避免左和右重尾巴的基于DNN的搜索方法。

Generative models have been widely applied to solve extractive tasks, where parts of the input is extracted to form the desired output, and achieved significant success. For example, in extractive question answering (QA), generative models have constantly yielded state-of-the-art results. In this work, we identify the issue of tokenization inconsistency that is commonly neglected in training these models. This issue damages the extractive nature of these tasks after the input and output are tokenized inconsistently by the tokenizer, and thus leads to performance drop as well as hallucination. We propose a simple yet effective fix to this issue and conduct a case study on extractive QA. We show that, with consistent tokenization, the model performs better in both in-domain and out-of-domain datasets, with a notable average of +1.7 F2 gain when a BART model is trained on SQuAD and evaluated on 8 QA datasets. Further, the model converges faster, and becomes less likely to generate out-of-context answers. With these findings, we would like to call for more attention on how tokenization should be done when solving extractive tasks and recommend applying consistent tokenization during training.

An optimal delivery of arguments is key to persuasion in any debate, both for humans and for AI systems. This requires the use of clear and fluent claims relevant to the given debate. Prior work has studied the automatic assessment of argument quality extensively. Yet, no approach actually improves the quality so far. Our work is the first step towards filling this gap. We propose the task of claim optimization: to rewrite argumentative claims to optimize their delivery. As an initial approach, we first generate a candidate set of optimized claims using a sequence-to-sequence model, such as BART, while taking into account contextual information. Our key idea is then to rerank generated candidates with respect to different quality metrics to find the best optimization. In automatic and human evaluation, we outperform different reranking baselines on an English corpus, improving 60% of all claims (worsening 16% only). Follow-up analyses reveal that, beyond copy editing, our approach often specifies claims with details, whereas it adds less evidence than humans do. Moreover, its capabilities generalize well to other domains, such as instructional texts.

Although pre-trained language models (PLMs) have shown impressive performance by text-only self-supervised training, they are found lack of visual semantics or commonsense, e.g., sizes, shapes, and colors of commonplace objects. Existing solutions often rely on explicit images for visual knowledge augmentation (requiring time-consuming retrieval or generation), and they also conduct the augmentation for the whole input text, without considering whether it is actually needed in specific inputs or tasks. To address these issues, we propose a novel visually-augmented fine-tuning approach that can be generally applied to various PLMs or NLP tasks, without using any retrieved or generated images, namely VAWI. Specifically, we first identify the visually-hungry words (VH-words) from input text via a token selector, where three different methods have been proposed, including syntax-, attention- and learning-based strategies. Then, we adopt a fixed CLIP text encoder to generate the visually-augmented representations of these VH-words. As it has been pre-trained by vision-language alignment task on the large-scale corpus, it is capable of injecting visual semantics into the aligned text representations. Finally, the visually-augmented features will be fused and transformed into the pre-designed visual prompts based on VH-words, which can be inserted into PLMs to enrich the visual semantics in word representations. We conduct extensive experiments on ten NLP tasks, i.e., GLUE benchmark, CommonsenseQA, CommonGen, and SNLI-VE. Experimental results show that our approach can consistently improve the performance of BERT, RoBERTa, BART, and T5 at different scales, and outperform several competitive baselines significantly. Our codes and data are publicly available at~\url{https://github.com/RUCAIBox/VAWI}.

We propose a combined three pre-trained language models (XLM-R, BART, and DeBERTa-V3) as an empower of contextualized embedding for named entity recognition. Our model achieves a 92.9% F1 score on the test set and ranks 5th on the leaderboard at NL4Opt competition subtask 1.

Diabetic retinopathy (DR) is a complication of diabetes, and one of the major causes of vision impairment in the global population. As the early-stage manifestation of DR is usually very mild and hard to detect, an accurate diagnosis via eye-screening is clinically important to prevent vision loss at later stages. In this work, we propose an ensemble method to automatically grade DR using ultra-wide optical coherence tomography angiography (UW-OCTA) images available from Diabetic Retinopathy Analysis Challenge (DRAC) 2022. First, we adopt the state-of-the-art classification networks, i.e., ResNet, DenseNet, EfficientNet, and VGG, and train them to grade UW-OCTA images with different splits of the available dataset. Ultimately, we obtain 25 models, of which, the top 16 models are selected and ensembled to generate the final predictions. During the training process, we also investigate the multi-task learning strategy, and add an auxiliary classification task, the Image Quality Assessment, to improve the model performance. Our final ensemble model achieved a quadratic weighted kappa (QWK) of 0.9346 and an Area Under Curve (AUC) of 0.9766 on the internal testing dataset, and the QWK of 0.839 and the AUC of 0.8978 on the DRAC challenge testing dataset.

Modern embedding-based metrics for evaluation of generated text generally fall into one of two paradigms: discriminative metrics that are trained to directly predict which outputs are of higher quality according to supervised human annotations, and generative metrics that are trained to evaluate text based on the probabilities of a generative model. Both have their advantages; discriminative metrics are able to directly optimize for the problem of distinguishing between good and bad outputs, while generative metrics can be trained using abundant raw text. In this paper, we present a framework that combines the best of both worlds, using both supervised and unsupervised signals from whatever data we have available. We operationalize this idea by training T5Score, a metric that uses these training signals with mT5 as the backbone. We perform an extensive empirical comparison with other existing metrics on 5 datasets, 19 languages and 280 systems, demonstrating the utility of our method. Experimental results show that: T5Score achieves the best performance on all datasets against existing top-scoring metrics at the segment level. We release our code and models at https://github.com/qinyiwei/T5Score.

A primary objective of news articles is to establish the factual record for an event, frequently achieved by conveying both the details of the specified event (i.e., the 5 Ws; Who, What, Where, When and Why regarding the event) and how people reacted to it (i.e., reported statements). However, existing work on news summarization almost exclusively focuses on the event details. In this work, we propose the novel task of summarizing the reactions of different speakers, as expressed by their reported statements, to a given event. To this end, we create a new multi-document summarization benchmark, SUMREN, comprising 745 summaries of reported statements from various public figures obtained from 633 news articles discussing 132 events. We propose an automatic silver training data generation approach for our task, which helps smaller models like BART achieve GPT-3 level performance on this task. Finally, we introduce a pipeline-based framework for summarizing reported speech, which we empirically show to generate summaries that are more abstractive and factual than baseline query-focused summarization approaches.

Approximation fixpoint theory (AFT) is an abstract and general algebraic framework for studying the semantics of nonmonotonic logics. It provides a unifying study of the semantics of different formalisms for nonmonotonic reasoning, such as logic programming, default logic and autoepistemic logic. In this paper, we extend AFT to dealing with non-deterministic constructs that allow to handle indefinite information, represented e.g. by disjunctive formulas. This is done by generalizing the main constructions and corresponding results of AFT to non-deterministic operators, whose ranges are sets of elements rather than single elements. The applicability and usefulness of this generalization is illustrated in the context of disjunctive logic programming.