As a promising paradigm to collaboratively train models with decentralized data, Federated Learning (FL) can be exploited to fine-tune Large Language Models (LLMs). While LLMs correspond to huge size, the scale of the training data significantly increases, which leads to tremendous amounts of computation and communication costs. The training data is generally non-Independent and Identically Distributed (non-IID), which requires adaptive data processing within each device. Although Low-Rank Adaptation (LoRA) can significantly reduce the scale of parameters to update in the fine-tuning process, it still takes unaffordable time to transfer the low-rank parameters of all the layers in LLMs. In this paper, we propose a Fisher Information-based Efficient Curriculum Federated Learning framework (FibecFed) with two novel methods, i.e., adaptive federated curriculum learning and efficient sparse parameter update. First, we propose a fisher information-based method to adaptively sample data within each device to improve the effectiveness of the FL fine-tuning process. Second, we dynamically select the proper layers for global aggregation and sparse parameters for local update with LoRA so as to improve the efficiency of the FL fine-tuning process. Extensive experimental results based on 10 datasets demonstrate that FibecFed yields excellent performance (up to 45.35% in terms of accuracy) and superb fine-tuning speed (up to 98.61% faster) compared with 17 baseline approaches).

Federated learning (FL) is a promising paradigm to enable collaborative model training with decentralized data. However, the training process of Large Language Models (LLMs) generally incurs the update of significant parameters, which limits the applicability of FL techniques to tackle the LLMs in real scenarios. Prompt tuning can significantly reduce the number of parameters to update, but it either incurs performance degradation or low training efficiency. The straightforward utilization of prompt tuning in the FL often raises non-trivial communication costs and dramatically degrades performance. In addition, the decentralized data is generally non-Independent and Identically Distributed (non-IID), which brings client drift problems and thus poor performance. This paper proposes a Parameter-efficient prompt Tuning approach with Adaptive Optimization, i.e., FedPepTAO, to enable efficient and effective FL of LLMs. First, an efficient partial prompt tuning approach is proposed to improve performance and efficiency simultaneously. Second, a novel adaptive optimization method is developed to address the client drift problems on both the device and server sides to enhance performance further. Extensive experiments based on 10 datasets demonstrate the superb performance (up to 60.8% in terms of accuracy) and efficiency (up to 97.59% in terms of training time) of FedPepTAO compared with 9 baseline approaches. Our code is available at https://github.com/llm-eff/FedPepTAO.

Although remarkable progress on the neural table-to-text methods has been made, the generalization issues hinder the applicability of these models due to the limited source tables. Large-scale pretrained language models sound like a promising solution to tackle such issues. However, how to effectively bridge the gap between the structured table and the text input by fully leveraging table information to fuel the pretrained model is still not well explored. Besides, another challenge of integrating the deliberation mechanism into the text-to-text pretrained model for solving the table-to-text task remains seldom studied. In this paper, to implement the table-to-text generation with pretrained language model, we propose a table structure understanding and text deliberating approach, namely TASD. To be specific, we devise a three-layered multi-head attention network to realize the table-structureaware text generation model with the help of the pretrained language model. Furthermore, a multi-pass decoder framework is adopted to enhance the capability of polishing generated text for table descriptions. The empirical studies, as well as human evaluation, on two public datasets, validate that our approach can generate faithful and fluent descriptive texts for different types of tables.

Short text classification (STC) is hard as short texts lack context information and labeled data is not enough. Graph neural networks obtain the state-of-the-art on STC since they can merge various auxiliary information via the message passing framework. However, existing works conduct transductive learning, which requires retraining to accommodate new samples and takes large memory. In this paper, we present SimpleSTC which handles inductive STC problem but only leverages words. We construct word graph from an external large corpus to compensate for the lack of semantic information, and learn text graph to handle the lack of labeled data. Results show that SimpleSTC obtains state-of-the-art performance with lower memory consumption and faster inference speed.

Entity linking (EL) on short text is crucial for a variety of industrial applications. Compared with general long-text EL, short-text EL poses particular challenges as the limited context restricts the clues one can leverage to disambiguate textual mentions. On the other hand, existing studies mostly focus on black-box neural methods and thus lack interpretability, which is critical to industrial applications in certain areas. In this study, we extend upon LNN-EL, a monolingual short-text EL method based on interpretable first-order logic, by incorporating three sets of multilingual features to enable disambiguating mentions written in languages other than English. More specifically, we use multilingual autoencoding language models (i.e., mBERT) to capture the similarities between the mention with its context and the candidate entity; we use multilingual sequence-to-sequence language models (i.e., mBART and mT5) to represent the likelihood of the text given the candidate entity. We also propose a word-level context feature to capture the semantic evidence of the co-occurring mentions. We evaluate the proposed xLNN-EL approach on the QALD-9-multilingual dataset and demonstrate the cross-linguality of the model and the effectiveness of the features.

Pre-trained language models (PLMs) can provide a good starting point for downstream applications. However, it is difficult to generalize PLMs to new tasks given a few labeled samples. In this work, we show that Relation Graph augmented Learning (RGL) can improve the performance of few-shot natural language understanding tasks. During learning, RGL constructs a relation graph based on the label consistency between samples in the same batch, and learns to solve the resultant node classification and link prediction problems on the relation graph. In this way, RGL fully exploits the limited supervised information, which can boost the tuning effectiveness. Extensive experimental results show that RGL consistently improves the performance of prompt-based tuning strategies.

In the past few years, large pre-trained language models (PLMs) have been widely adopted in different areas and have made fundamental improvements over a variety of downstream tasks in natural language processing (NLP). Meanwhile, domain-specific variants of PLMs are being proposed to address the needs of domains that demonstrate a specific pattern of writing and vocabulary, e.g., BioBERT for the biomedical domain and ClinicalBERT for the clinical domain. Recently, generative language models like BART and T5 are gaining popularity with their competitive performance on text generation as well as on tasks cast as generative problems. However, in the clinical domain, such domain-specific generative variants are still underexplored. To address this need, our work introduces a T5-based text-to-text transformer model pre-trained on clinical text, i.e., ClinicalT5. We evaluate the proposed model both intrinsically and extrinsically over a diverse set of tasks across multiple datasets, and show that ClinicalT5 dramatically outperforms T5 in the domain-specific tasks and compares favorably with its close baselines.

Fine-tuning pre-trained language models suchas BERT has become a common practice dom-inating leaderboards across various NLP tasks. Despite its recent success and wide adoption,this process is unstable when there are onlya small number of training samples available. The brittleness of this process is often reflectedby the sensitivity to random seeds. In this pa-per, we propose to tackle this problem basedon the noise stability property of deep nets,which is investigated in recent literature (Aroraet al., 2018; Sanyal et al., 2020). Specifically,we introduce a novel and effective regulariza-tion method to improve fine-tuning on NLPtasks, referred to asLayer-wiseNoiseStabilityRegularization (LNSR). We extend the theo-ries about adding noise to the input and provethat our method gives a stabler regularizationeffect. We provide supportive evidence by ex-perimentally confirming that well-performingmodels show a low sensitivity to noise andfine-tuning with LNSR exhibits clearly bet-ter generalizability and stability. Furthermore,our method also demonstrates advantages overother state-of-the-art algorithms including L2-SP (Li et al., 2018), Mixout (Lee et al., 2020)and SMART (Jiang et al., 20)

Social networks face a major challenge in the form of rumors and fake news, due to their intrinsic nature of connecting users to millions of others, and of giving any individual the power to post anything. Given the rapid, widespread dissemination of information in social networks, manually detecting suspicious news is sub-optimal. Thus, research on automatic rumor detection has become a necessity. Previous works in the domain have utilized the reply relations between posts, as well as the semantic similarity between the main post and its context, consisting of replies, in order to obtain state-of-the-art performance. In this work, we demonstrate that semantic oppositeness can improve the performance on the task of rumor detection. We show that semantic oppositeness captures elements of discord, which are not properly covered by previous efforts, which only utilize semantic similarity or reply structure. We show, with extensive experiments on recent data sets for this problem, that our proposed model achieves state-of-the-art performance. Further, we show that our model is more resistant to the variances in performance introduced by randomness.

Domain-specific pre-trained language models (PLMs) have achieved great success over various downstream tasks in different domains. However, existing domain-specific PLMs mostly rely on self-supervised learning over large amounts of domain text, without explicitly integrating domain-specific knowledge, which can be essential in many domains. Moreover, in knowledge-sensitive areas such as the biomedical domain, knowledge is stored in multiple sources and formats, and existing biomedical PLMs either neglect them or utilize them in a limited manner. In this work, we introduce an architecture to integrate domain knowledge from diverse sources into PLMs in a parameter-efficient way. More specifically, we propose to encode domain knowledge via adapters, which are small bottleneck feed-forward networks inserted between intermediate transformer layers in PLMs. These knowledge adapters are pre-trained for individual domain knowledge sources and integrated via an attention-based knowledge controller to enrich PLMs. Taking the biomedical domain as a case study, we explore three knowledge-specific adapters for PLMs based on the UMLS Metathesaurus graph, the Wikipedia articles for diseases, and the semantic grouping information for biomedical concepts. Extensive experiments on different biomedical NLP tasks and datasets demonstrate the benefits of the proposed architecture and the knowledge-specific adapters across multiple PLMs.

Short text classification is a fundamental task in natural language processing. It is hard due to the lack of context information and labeled data in practice. In this paper, we propose a new method called SHINE, which is based on graph neural network (GNN), for short text classification. First, we model the short text dataset as a hierarchical heterogeneous graph consisting of word-level component graphs which introduce more semantic and syntactic information. Then, we dynamically learn a short document graph that facilitates effective label propagation among similar short texts. Thus, comparing with existing GNN-based methods, SHINE can better exploit interactions between nodes of the same types and capture similarities between short texts. Extensive experiments on various benchmark short text datasets show that SHINE consistently outperforms state-of-the-art methods, especially with fewer labels.

Recent literatures have shown that knowledge graph (KG) learning models are highly vulnerable to adversarial attacks. However, there is still a paucity of vulnerability analyses of cross-lingual entity alignment under adversarial attacks. This paper proposes an adversarial attack model with two novel attack techniques to perturb the KG structure and degrade the quality of deep cross-lingual entity alignment. First, an entity density maximization method is employed to hide the attacked entities in dense regions in two KGs, such that the derived perturbations are unnoticeable. Second, an attack signal amplification method is developed to reduce the gradient vanishing issues in the process of adversarial attacks for further improving the attack effectiveness.

Network representation learning (NRL) is crucial in the area of graph learning. Recently, graph autoencoders and its variants have gained much attention and popularity among various types of node embedding approaches. Most existing graph autoencoder-based methods aim to minimize the reconstruction errors of the input network while not explicitly considering the semantic relatedness between nodes. In this paper, we propose a novel network embedding method which models the consistency across different views of networks. More specifically, we create a second view from the input network which captures the relation between nodes based on node content and enforce the latent representations from the two views to be consistent by incorporating a multiview adversarial regularization module. The experimental studies on benchmark datasets prove the effectiveness of this method, and demonstrate that our method compares favorably with the state-of-the-art algorithms on challenging tasks such as link prediction and node clustering. We also evaluate our method on a real-world application, i.e., 30-day unplanned ICU readmission prediction, and achieve promising results compared with several baseline methods.

This paper studies the task of Relation Extraction (RE) that aims to identify the semantic relations between two entity mentions in text. In the deep learning models for RE, it has been beneficial to incorporate the syntactic structures from the dependency trees of the input sentences. In such models, the dependency trees are often used to directly structure the network architectures or to obtain the dependency relations between the word pairs to inject the syntactic information into the models via multi-task learning. The major problem with these approaches is the lack of generalization beyond the syntactic structures in the training data or the failure to capture the syntactic importance of the words for RE. In order to overcome these issues, we propose a novel deep learning model for RE that uses the dependency trees to extract the syntax-based importance scores for the words, serving as a tree representation to introduce syntactic information into the models with greater generalization. In particular, we leverage Ordered-Neuron Long-Short Term Memory Networks (ON-LSTM) to infer the model-based importance scores for RE for every word in the sentences that are then regulated to be consistent with the syntax-based scores to enable syntactic information injection. We perform extensive experiments to demonstrate the effectiveness of the proposed method, leading to the state-of-the-art performance on three RE benchmark datasets.

Aspect-based Sentiment Analysis (ABSA) seeks to predict the sentiment polarity of a sentence toward a specific aspect. Recently, it has been shown that dependency trees can be integrated into deep learning models to produce the state-of-the-art performance for ABSA. However, these models tend to compute the hidden/representation vectors without considering the aspect terms and fail to benefit from the overall contextual importance scores of the words that can be obtained from the dependency tree for ABSA. In this work, we propose a novel graph-based deep learning model to overcome these two issues of the prior work on ABSA. In our model, gate vectors are generated from the representation vectors of the aspect terms to customize the hidden vectors of the graph-based models toward the aspect terms. In addition, we propose a mechanism to obtain the importance scores for each word in the sentences based on the dependency trees that are then injected into the model to improve the representation vectors for ABSA. The proposed model achieves the state-of-the-art performance on three benchmark datasets.

Targeted opinion word extraction (TOWE) is a sub-task of aspect based sentiment analysis (ABSA) which aims to find the opinion words for a given aspect-term in a sentence. Despite their success for TOWE, the current deep learning models fail to exploit the syntactic information of the sentences that have been proved to be useful for TOWE in the prior research. In this work, we propose to incorporate the syntactic structures of the sentences into the deep learning models for TOWE, leveraging the syntax-based opinion possibility scores and the syntactic connections between the words. We also introduce a novel regularization technique to improve the performance of the deep learning models based on the representation distinctions between the words in TOWE. The proposed model is extensively analyzed and achieves the state-of-the-art performance on four benchmark datasets.

Unsupervised word embeddings have become a popular approach of word representation in NLP tasks. However there are limitations to the semantics represented by unsupervised embeddings, and inadequate fine-tuning of embeddings can lead to suboptimal performance. We propose a novel learning technique called Delta Embedding Learning, which can be applied to general NLP tasks to improve performance by optimized tuning of the word embeddings. A structured regularization is applied to the embeddings to ensure they are tuned in an incremental way. As a result, the tuned word embeddings become better word representations by absorbing semantic information from supervision without “forgetting.” We apply the method to various NLP tasks and see a consistent improvement in performance. Evaluation also confirms the tuned word embeddings have better semantic properties.

Event factuality prediction (EFP) is the task of assessing the degree to which an event mentioned in a sentence has happened. For this task, both syntactic and semantic information are crucial to identify the important context words. The previous work for EFP has only combined these information in a simple way that cannot fully exploit their coordination. In this work, we introduce a novel graph-based neural network for EFP that can integrate the semantic and syntactic information more effectively. Our experiments demonstrate the advantage of the proposed model for EFP.

We propose an efficient method to generate white-box adversarial examples to trick a character-level neural classifier. We find that only a few manipulations are needed to greatly decrease the accuracy. Our method relies on an atomic flip operation, which swaps one token for another, based on the gradients of the one-hot input vectors. Due to efficiency of our method, we can perform adversarial training which makes the model more robust to attacks at test time. With the use of a few semantics-preserving constraints, we demonstrate that HotFlip can be adapted to attack a word-level classifier as well.

Evaluating on adversarial examples has become a standard procedure to measure robustness of deep learning models. Due to the difficulty of creating white-box adversarial examples for discrete text input, most analyses of the robustness of NLP models have been done through black-box adversarial examples. We investigate adversarial examples for character-level neural machine translation (NMT), and contrast black-box adversaries with a novel white-box adversary, which employs differentiable string-edit operations to rank adversarial changes. We propose two novel types of attacks which aim to remove or change a word in a translation, rather than simply break the NMT. We demonstrate that white-box adversarial examples are significantly stronger than their black-box counterparts in different attack scenarios, which show more serious vulnerabilities than previously known. In addition, after performing adversarial training, which takes only 3 times longer than regular training, we can improve the model’s robustness significantly.

Classifying the stance expressed in online microblogging social media is an emerging problem in opinion mining. We propose a probabilistic approach to stance classification in tweets, which models stance, target of stance, and sentiment of tweet, jointly. Instead of simply conjoining the sentiment or target variables as extra variables to the feature space, we use a novel formulation to incorporate three-way interactions among sentiment-stance-input variables and three-way interactions among target-stance-input variables. The proposed specification intuitively aims to discriminate sentiment features from target features for stance classification. In addition, regularizing a single stance classifier, which handles all targets, acts as a soft weight-sharing among them. We demonstrate that discriminative training of this model achieves the state-of-the-art results in supervised stance classification, and its generative training obtains competitive results in the weakly supervised setting.