The increasing size and complexity of large language models (LLMs) have raised significant challenges in deployment efficiency, particularly under resource constraints. Post-training quantization (PTQ) has emerged as a practical solution by compressing models without requiring retraining. While existing methods focus on uniform quantization schemes for both weights and activations, they often suffer from substantial accuracy degradation due to the non-uniform nature of activation distributions. In this work, we revisit the activation quantization problem from a theoretical perspective grounded in the Lloyd-Max optimality conditions. We identify the core issue as the non-uniform distribution of activations within the quantization interval, which causes the optimal quantization point under the Lloyd-Max criterion to shift away from the midpoint of the interval. To address this issue, we propose a two-stage orthogonal transformation method, RUQuant. In the first stage, activations are divided into blocks. Each block is mapped to uniformly sampled target vectors using composite orthogonal matrices, which are constructed from Householder reflections and Givens rotations. In the second stage, a global Householder reflection is fine-tuned to further minimize quantization error using Transformer output discrepancies. Empirical results show that our method achieves near-optimal quantization performance without requiring model fine-tuning: RUQuant achieves 99.8% of full-precision accuracy with W6A6 and 97% with W4A4 quantization for a 13B LLM, within approximately one minute. A fine-tuned variant yields even higher accuracy, demonstrating the effectiveness and scalability of our approach.

Multi-label few-shot image classification is a crucial and challenging task due to limited annotated data and elusive category specificity. However, research on this topic is still in the rudimentary stage and few methods are available. Existing methods either leverage data augmentation to alleviate data scarcity or utilize label features as auxiliary knowledge to eliminate the negative effect caused by irrelevant categories, but they ignore the utilization of image region features for data augmentation, and overlook to learn appropriate text feature to better match the image features of specific categories. Moreover, these methods only focus on one side and do not effectively tackle the above two issues simultaneously. In this paper, we introduce a novel prototype-based multi-label few-shot learning framework that seamlessly integrates pairwise feature augmentation and flexible prompt learning. Specifically, by pairwise feature augmentation, we leverage the region features of images in the support set to generate more image features and construct image prototypes, thus alleviating the issue of data scarcity. By flexible prompt learning, we adaptively acquire class-specific prompts to build text prototypes that highly match the image features of specific classes, thereby mitigating the impact of irrelevant classes. Finally, with adaptive learnable parameters, we merge image and text prototypes to obtain the final prototypes, achieving a more powerful classifier for multi-label few-shot image classification. Extensive experimental results demonstrate that our proposed method can push the performance to a higher level.

With the continuously expanding parameters, efficiently adapting large language models to downstream tasks is crucial in resource-limited conditions.Many parameter-efficient finetuning methods have emerged to address this challenge.However, they lack flexibility, like LoRA requires manually selecting trainable parameters and rank size, (IA) 3 can only scale the activations along columns, yielding inferior results due to less precise fine-tuning.To address these issues, we propose a novel method named AdaDHP with fewer parameters and finer granularity, which can adaptively select important parameters for each task.Specifically, we introduce two trainable vectors for each parameter and fine-tune the parameters through Hadamard product along both rows and columns.This significantly reduces the number of trainable parameters, with our parameter count capped at the lower limit of LoRA.Moreover, we design an adaptive parameter selection strategy to select important parameters for downstream tasks dynamically.This allows our method to flexibly remove unimportant parameters for downstream tasks.Finally, we demonstrate the superiority of our method on the T5-base model across 17 NLU tasks and on complex mathematical tasks with the Llama series models.

Most current image captioning models rely on the autoregressive approach, which unfortunately results in significant inference delays that hinder their practical use. In contrast, non-autoregressive methods show promising potential for increasing inference speeds. However, there is often a performance gap between the non-autoregressive and autoregressive image captioning models due to issues like word repetition and semantic inconsistencies in the generated captions. Autoregressive models benefit from self-critical sequence training, which helps produce more coherent and fluid captions. While non-autoregressive models are difficult to benefit from as they predict words independently. In this paper, we introduce a two-stage training strategy designed to harness self-critical sequence training for enhancing the non-autoregressive image captioning model. Our approach initially treats the image captioning task as a multi-label classification problem, which allows for the stable production of multiple candidate captions. In the second stage, we employ these candidate captions to compute sequence-level evaluation metric scores that serve as reward scores for self-critical sequence training. Extensive experiments demonstrate the effectiveness of our proposed method and show that our model achieves a new state-of-the-art performance in inference accuracy and speed.

Large language models (LLMs) have shown remarkable performance in various domains, but they are constrained by massive computational and storage costs. Quantization, an effective technique for compressing models to fit resource-limited devices while preserving generative quality, encompasses two primary methods: quantization aware training (QAT) and post-training quantization (PTQ). QAT involves additional retraining or fine-tuning, thus inevitably resulting in high training cost and making it unsuitable for LLMs. Consequently, PTQ has become the research hotspot in recent quantization methods. However, existing PTQ methods usually rely on various complex computation procedures and suffer from considerable performance degradation under low-bit quantization settings. To alleviate the above issues, we propose a simple and effective post-training quantization paradigm for LLMs, named SEPTQ. Specifically, SEPTQ first calculates the importance score for each element in the weight matrix and determines the quantization locations in a static global manner. Then it utilizes the mask matrix which represents the important locations to quantize and update the associated weights column-by-column until the appropriate quantized weight matrix is obtained. Compared with previous methods, SEPTQ simplifies the post-training quantization procedure into only two steps, and considers the effectiveness and efficiency simultaneously. Experimental results on various datasets across a suite of models ranging from millions to billions in different quantization bit-levels demonstrate that SEPTQ significantly outperforms other strong baselines, especially in low-bit quantization scenarios.

Black-box adversarial attack on vision-language pre-trained models is a practical and challenging task, as text and image perturbations need to be considered simultaneously, and only the predicted results are accessible. Research on this problem is in its infancy, and only a handful of methods are available. Nevertheless, existing methods either rely on a complex iterative cross-search strategy, which inevitably consumes numerous queries, or only consider reducing the similarity of positive image-text pairs but ignore that of negative ones, which will also be implicitly diminished, thus inevitably affecting the attack performance. To alleviate the above issues, we propose a simple yet effective framework to generate high-quality adversarial examples on vision-language pre-trained models, named HQA-VLAttack, which consists of text and image attack stages. For text perturbation generation, it leverages the counter-fitting word vector to generate the substitute word set, thus guaranteeing the semantic consistency between the substitute word and the original word. For image perturbation generation, it first initializes the image adversarial example via the layer-importance guided strategy, and then utilizes contrastive learning to optimize the image adversarial perturbation, which ensures that the similarity of positive image-text pairs is decreased while that of negative image-text pairs is increased. In this way, the optimized adversarial images and texts are more likely to retrieve negative examples, thereby enhancing the attack success rate. Experimental results on three benchmark datasets demonstrate that HQA-VLAttack significantly outperforms strong baselines in terms of attack success rate.

The emergence of the Intelligent Internet of Things (IIoT) has brought data processing closer to data sources, especially for real-time processing of surveillance video and image analysis. Image captioning plays a crucial role in understanding images. However, the Transformer architecture, which has become prevalent in recent applications, has been observed to increase the computational resources required for image captioning models. Conventionally, most existing methods use the autoregressive paradigm, which reduces their computational efficiency on edge devices and results in significant inference delays. In this paper, we use non-autoregressive paradigms to improve its inference speed and model efficiency. Nevertheless, the lack of effective inputs results in a performance gap between non-autoregressive and autoregressive models. To bridge this gap, we propose the learnable positional encoding and keyword guided non-autoregressive image captioning. Firstly, a diffusion model guided by image features is employed to generate keywords that accurately reflect the image content, thereby infusing a substantial amount of semantic information into the non-autoregressive decoder. Secondly, positional encoding is utilized to guide the decoder in generating appropriate words at the correct positions within the caption. Extensive experiments on widely used benchmarks demonstrate that our model achieves state-of-the-art performance in non-autoregressive image captioning. Furthermore, our model maintains a competitive inference speed.

Depression detection is a challenging and crucial task in psychological illness diagnosis. Utilizing online user posts to predict whether a user suffers from depression seems an effective and promising direction. However, existing methods suffer from either poor interpretability brought by the black-box models or underwhelming performance caused by the completely separate two-stage model structure. To alleviate these limitations, we propose a novel capsule network integrated with contrastive learning for depression detection (DeCapsNet). The highlights of DeCapsNet can be summarized as follows. First, it extracts symptom capsules from user posts by leveraging meticulously designed symptom descriptions, and then distills them into class-indicative depression capsules. The overall workflow is in an explicit hierarchical reasoning manner and can be well interpreted by the Patient Health Questionnaire-9 (PHQ9), which is one of the most widely adopted questionnaires for depression diagnosis. Second, it integrates with contrastive learning, which can facilitate the embeddings from the same class to be pulled closer, while simultaneously pushing the embeddings from different classes apart. In addition, by adopting the end-to-end training strategy, it does not necessitate additional data annotation, and mitigates the potential adverse effects from the upstream task to the downstream task. Extensive experiments on three widely-used datasets show that in both within-dataset and cross-dataset scenarios our proposed method outperforms other strong baselines significantly.

Few-shot and zero-shot text classification aim to recognize samples from novel classes with limited labeled samples or no labeled samples at all. While prevailing methods have shown promising performance via transferring knowledge from seen classes to unseen classes, they are still limited by (1) Inherent dissimilarities among classes make the transformation of features learned from seen classes to unseen classes both difficult and inefficient. (2) Rare labeled novel samples usually cannot provide enough supervision signals to enable the model to adjust from the source distribution to the target distribution, especially for complicated scenarios. To alleviate the above issues, we propose a simple and effective strategy for few-shot and zero-shot text classification. We aim to liberate the model from the confines of seen classes, thereby enabling it to predict unseen categories without the necessity of training on seen classes. Specifically, for mining more related unseen category knowledge, we utilize a large pre-trained language model to generate pseudo novel samples, and select the most representative ones as category anchors. After that, we convert the multi-class classification task into a binary classification task and use the similarities of query-anchor pairs for prediction to fully leverage the limited supervision signals. Extensive experiments on six widely used public datasets show that our proposed method can outperform other strong baselines significantly in few-shot and zero-shot tasks, even without using any seen class samples.

Multi-goal conversational recommender system (MG-CRS) which is more in line with realistic scenarios has attracted a lot of attention. MG-CRS can dynamically capture the demands of users in conversation, continuously engage their interests, and make recommendations. The key of accomplishing these tasks is to plan a reasonable goal sequence which can naturally guide the user to accept the recommended goal. Previous works have demonstrated that mining the correlations of goals from the goal sequences in the dialogue corpus is helpful for recommending the goal that the user is interested in. However, they independently model correlations for each level of goal (i.e., goal type or entity) and neglect the order of goals appear in the dialogue. In this paper, we propose a goal interaction graph planning framework which constructs a directed heterogeneous graph to flexibly model the correlations between any level of goals and retain the order of goals. We design a goal interaction graph learning module to model the goal correlations and propagate goal representations via directed edges, then use an encoder and a dual-way fusion decoder to extract the most relevant information with the current goal from the conversation and domain knowledge, making the next-goal prediction fully exploit the prior goal correlations and user feedback. Finally we generate engaging responses based on the predicted goal sequence to complete the recommendation task. Experiments on two benchmark datasets show that our method achieves significant improvements in both the goal planning and response generation tasks.

New intent discovery aims to identify new intents from unlabeled utterances in the open-world scenario. As the fundamental and challenging problem in dialogue systems, new intent discovery attracts increasing attention but is still under exploration. In this paper, we propose a simple and effective new intent discovery framework with multi-view clustering. Specifically, we first adopt a double-branch representation learning strategy to learn high-quality utterance representations. Then we conduct a multi-view clustering method to obtain the satisfactory cluster assignment in an iterative manner, thus fulfilling the new intent discovery task. Extensive experimental results on three widely-used datasets demonstrate that our proposed method outperforms other strong baselines in most cases.

Catastrophic forgetting has become an intractable problem in the continual learning setting because previous data is not accessible when training. To mitigate this problem, memory-based continual learning methods replay previous data from a fixed-size memory buffer. Reservoir sampling, which can sample uniformly from streaming data in a single pass and randomly delete a few buffered samples to keep the memory capacity, is widely used in the memory update process. Nevertheless, the random deletion will cause the agent to miss some high-quality samples, which are more effective in overcoming forgetting. In this paper, we propose a novel memory update method, Class-level Minimally Interfered Update (CMIU), which can maintain the sampling randomness and remove the worthless samples from the same classes as the randomly deleting samples. Concretely, CMIU deletes the samples that are minimally interfered with the agent update. The experiments show that CMIU presents improved performance compared to the baselines on both balanced and imbalanced data streams in the online continual learning setting.

Multi-label intent detection aims to recognize multiple user intents behind dialogue utterances. The diversity of user utterances and the scarcity of training data motivate multi-label few-shot intent detection. However, existing methods ignore the hybrid of verb and noun within an intent, which is essential to identify the user intent. In this paper, we propose a label hierarchical structure-aware method for multi-label few-shot intent detection via prompt tuning (LHS). Firstly, for the support data, we concatenate the original utterance with the label description generated by GPT-4 to obtain the utterance-level representation. Then we construct a multi-label hierarchical structure-aware prompt model to learn the label hierarchical information. To learn more discriminative class prototypes, we devise a prototypical contrastive learning method to pull the utterances close to their corresponding intent labels and away from other intent labels. Extensive experiments on two datasets demonstrate the superiority of our method.

Hard-label textual adversarial attack is a challenging task, as only the predicted label information is available, and the text space is discrete and non-differentiable. Relevant research work is still in fancy and just a handful of methods are proposed. However, existing methods suffer from either the high complexity of genetic algorithms or inaccurate gradient estimation, thus are arduous to obtain adversarial examples with high semantic similarity and low perturbation rate under the tight-budget scenario. In this paper, we propose a simple and sweet paradigm for hard-label textual adversarial attack, named SSPAttack. Specifically, SSPAttack first utilizes initialization to generate an adversarial example, and removes unnecessary replacement words to reduce the number of changed words. Then it determines the replacement order and searches for an anchor synonym, thus avoiding going through all the synonyms. Finally, it pushes substitution words towards original words until an appropriate adversarial example is obtained. The core idea of SSPAttack is just swapping words whose mechanism is simple. Experimental results on eight benchmark datasets and two real-world APIs have shown that the performance of SSPAttack is sweet in terms of similarity, perturbation rate and query efficiency.

Distribution estimation has been demonstrated as one of the most effective approaches in dealing with few-shot image classification, as the low-level patterns and underlying representations can be easily transferred across different tasks in computer vision domain. However, directly applying this approach to few-shot text classification is challenging, since leveraging the statistics of known classes with sufficient samples to calibrate the distributions of novel classes may cause negative effects due to serious category difference in text domain. To alleviate this issue, we propose two simple yet effective strategies to estimate the distributions of the novel classes by utilizing unlabeled query samples, thus avoiding the potential negative transfer issue. Specifically, we first assume a class or sample follows the Gaussian distribution, and use the original support set and the nearest few query samples to estimate the corresponding mean and covariance. Then, we augment the labeled samples by sampling from the estimated distribution, which can provide sufficient supervision for training the classification model. Extensive experiments on eight few-shot text classification datasets show that the proposed method outperforms state-of-the-art baselines significantly.

Meta-learning methods have shown to be effective in dealing with cold-start recommendation. However, most previous methods rely on an ideal assumption that there exists a similar data distribution between source and target tasks, which are unsuitable for the scenario that only extremely limited number of new user or item interactions are available. In this paper, we propose to boost meta-learning cold-start recommendation with graph neural network (MeGNN). First, it utilizes the global neighborhood translation learning to obtain consistent potential interactions for all new user and item nodes, which can refine their representations. Second, it employs the local neighborhood translation learning to predict specific potential interactions for each node, thus guaranteeing the personalized requirement. In experiments, we combine MeGNN with two representative meta-learning models MeLU and TaNP. Extensive results on two widely-used datasets show the superiority of MeGNN in four different scenarios.

Emotion recognition in conversation (ERC) enables a deeper understanding of emotion for each utterance within a conversation. Recent progress on ERC has proved that using Graph Neural Networks (GNN) to model conversational context is effective for identifying emotions. However, existing GNN-based approaches still suffer from two limitations: (1) they model the context of each utterance with a certain window, which ignores the diversity of emotion changes of utterances in conversation; (2) they mostly take no account of additional knowledge information, which limits the performance of ERC. In this paper, we propose a knowledge-aware graph convo-lutional network (KGCN-ERC) by introducing a knowledge graph into node connection of graph neural networks for the first time. Based on the rich sentiment knowledge, KGCN-ERC searches for the most appropriate local window for each utterance and builds sensible utterance connections. Experiments show that our approach achieves competitive performance compared with state-of-the-art ERC methods.

A class-incremental learning agent learns online with a neverending stream of data in one training epoch. In this setting, the agent suffers from severe catastrophic forgetting due to the absence of data from the observed classes after learning data from new classes. Besides, the prototypes rapidly become outdated as the agent adapts to new data sequentially, and the previous example embeddings spread out in an unforeseen way, which exacerbates forgetting (i.e., concept drift). Based on this observation, we propose a replay-based method, called TriCL, which gathers the embeddings near the prototype from the same class and separates the embeddings from the different class prototypes. TriCL leverages an improved triplet loss without extra arranged input data triplets. To facilitate rapid convergence between the same class samples, we design a memory update algorithm for decreasing the variance among the buffered samples from the same class. Furthermore, we make a prototype compensation strategy for preventing drift. Compared to the state-of-the-art benchmarks, the experiments demonstrate that our proposed method presents improved performance in the online class-incremental learning setting.

Decision-based methods have shown to be effective in black-box adversarial attacks, as they can obtain satisfactory performance and only require to access the final model prediction. Gradient estimation is a critical step in black-box adversarial attacks, as it will directly affect the query efficiency. Recent works have attempted to utilize gradient priors to facilitate score-based methods to obtain better results. However, these gradient priors still suffer from the edge gradient discrepancy issue and the successive iteration gradient direction issue, thus are difficult to simply extend to decision-based methods. In this paper, we propose a novel Decision-based Black-box Attack framework with Gradient Priors (DBA-GP), which seamlessly integrates the data-dependent gradient prior and time-dependent prior into the gradient estimation procedure. First, by leveraging the joint bilateral filter to deal with each random perturbation, DBA-GP can guarantee that the generated perturbations in edge locations are hardly smoothed, i.e., alleviating the edge gradient discrepancy, thus remaining the characteristics of the original image as much as possible. Second, by utilizing a new gradient updating strategy to automatically adjust the successive iteration gradient direction, DBA-GP can accelerate the convergence speed, thus improving the query efficiency. Extensive experiments have demonstrated that the proposed method outperforms other strong baselines significantly.

Black-box hard-label adversarial attack on text is a practical and challenging task, as the text data space is inherently discrete and non-differentiable, and only the predicted label is accessible. Research on this problem is still in the embryonic stage and only a few methods are available. Nevertheless, existing methods rely on the complex heuristic algorithm or unreliable gradient estimation strategy, which probably fall into the local optimum and inevitably consume numerous queries, thus are difficult to craft satisfactory adversarial examples with high semantic similarity and low perturbation rate in a limited query budget. To alleviate above issues, we propose a simple yet effective framework to generate high quality textual adversarial examples under the black-box hard-label attack scenarios, named HQA-Attack. Specifically, after initializing an adversarial example randomly, HQA-attack first constantly substitutes original words back as many as possible, thus shrinking the perturbation rate. Then it leverages the synonym set of the remaining changed words to further optimize the adversarial example with the direction which can improve the semantic similarity and satisfy the adversarial condition simultaneously. In addition, during the optimizing procedure, it searches a transition synonym word for each changed word, thus avoiding traversing the whole synonym set and reducing the query number to some extent. Extensive experimental results on five text classification datasets, three natural language inference datasets and two real-world APIs have shown that the proposed HQA-Attack method outperforms other strong baselines significantly.

Attributed graph clustering is a challenging task as it requires to jointly model graph structure and node attributes. Although recent advances in graph convolutional networks have shown the effectiveness of graph convolution in combining structural and content information, there is limited understanding of how to properly apply it for attributed graph clustering. Previous methods commonly use a fixed and low order graph convolution, which only aggregates information of few-hop neighbours and hence cannot fully capture the cluster structures of diverse graphs. In this paper, we first propose an adaptive graph convolution method (AGC) for attributed graph clustering, which exploits high-order graph convolutions to capture global cluster structures and adaptively selects an appropriate order$k$via intra-cluster distance. While AGC can find a reasonable$k$and avoid over-smoothing, it is not sensitive to the gradual decline of clustering performance as$k$increases. To search for a better$k$, we further propose an improved adaptive graph convolution method (IAGC) that not only observes the variation of intra-cluster distance, but also considers the inconsistencies of filtered features with graph structure and raw features, respectively. We establish the validity of our methods by theoretical analysis and extensive experiments on various benchmark datasets.

Data augmentation methods have proven highly effective in supervised learning domains where semantic-invariant perturbations can be easily applied to the labeled input examples. However, some used in deep reinforcement learning cannot guarantee the consistent semantics in the transition samples, leading to wrong value estimation and optimization during training. In this paper, we focus on the symmetric consistency in states and actions, and propose the Symmetric DQN to maintain it for better performance and data efficiency on small data regimes. Specifically, we adopt a consistent state flip and action projection on the original interaction transitions to construct the corresponding symmetric ones. The symmetric ones have similar middle rewards and terminal signals as the original ones but reverse states and actions. Therefore, by optimizing both the original and the symmetric losses, Symmetric DQN facilitates the value estimation from two directions per transition. We illustrate the promise of Symmetric DQN by conducting experiments on the Atari 100K benchmark. Symmetric DQN achieves a median HNS of 27.6% and a mean HNS of 51.1%, which surpasses most of the previous deep reinforcement learning methods with data augmentation.

Multi-label aspect category detection allows a given review sentence to contain multiple aspect categories, which is shown to be more practical in sentiment analysis and attracting increasing attention. As annotating large amounts of data is time-consuming and labor-intensive, data scarcity occurs frequently in real-world scenarios, which motivates multi-label few-shot aspect category detection. However, research on this problem is still in infancy and few methods are available. In this paper, we propose a novel label-enhanced prototypical network (LPN) for multi-label few-shot aspect category detection. The highlights of LPN can be summarized as follows. First, it leverages label description as auxiliary knowledge to learn more discriminative prototypes, which can retain aspect-relevant information while eliminating the harmful effect caused by irrelevant aspects. Second, it integrates with contrastive learning, which encourages that the sentences with the same aspect label are pulled together in embedding space while simultaneously pushing apart the sentences with different aspect labels. In addition, it introduces an adaptive multi-label inference module to predict the aspect count in the sentence, which is simple yet effective. Extensive experimental results on three datasets demonstrate that our proposed model LPN can consistently achieve state-of-the-art performance.

Zero-shot intent classification is a vital and challenging task in dialogue systems, which aims to deal with numerous fast-emerging unacquainted intents without annotated training data. To obtain more satisfactory performance, the crucial points lie in two aspects: extracting better utterance features and strengthening the model generalization ability. In this paper, we propose a simple yet effective meta-learning paradigm for zero-shot intent classification. To learn better semantic representations for utterances, we introduce a new mixture attention mechanism, which encodes the pertinent word occurrence patterns by leveraging the distributional signature attention and multi-layer perceptron attention simultaneously. To strengthen the transfer ability of the model from seen classes to unseen classes, we reformulate zero-shot intent classification with a meta-learning strategy, which trains the model by simulating multiple zero-shot classification tasks on seen categories, and promotes the model generalization ability with a meta-adapting procedure on mimic unseen categories. Extensive experiments on two real-world dialogue datasets in different languages show that our model outperforms other strong baselines on both standard and generalized zero-shot intent classification tasks.

User preference modeling is a vital yet challenging problem in personalized product search. In recent years, latent space based methods have achieved state-of-the-art performance by jointly learning semantic representations of products, users, and text tokens. However, existing methods are limited in their ability to model user preferences. They typically represent users by the products they visited in a short span of time using attentive models and lack the ability to exploit relational information such as user-product interactions or item co-occurrence relations. In this work, we propose to address the limitations of prior arts by exploring local and global user behavior patterns on a user successive behavior graph, which is constructed by utilizing short-term actions of all users. To capture implicit user preference signals and collaborative patterns, we use an efficient jumping graph convolution to explore high-order relations to enrich product representations for user preference modeling. Our approach can be seamlessly integrated with existing latent space based methods and be potentially applied in any product retrieval method that uses purchase history to model user preferences. Extensive experiments on eight Amazon benchmarks demonstrate the effectiveness and potential of our approach. The source code is available at https://github.com/floatSDSDS/SBG .

Graph clustering, which aims to partition a set of graphs into groups with similar structures, is a fundamental task in data analysis. With the great advances made by deep learning, deep graph clustering methods have achieved success. However, these methods have two limitations: (1) they learn graph embeddings by a neural language model that fails to effectively express graph properties, and (2) they treat embedding learning and clustering as two isolated processes, so the learned embeddings are unsuitable for the subsequent clustering. To overcome these limitations, we propose a novel capsule-based graph clustering (CGC) algorithm to cluster graphs. First, we construct a graph clustering capsule network (GCCN) that introduces capsules to capture graph properties. Second, we design an iterative optimization strategy to alternately update the GCCN parameters and clustering assignment parameters. This strategy leads GCCN to learn cluster-oriented graph embeddings. Experimental results show that our algorithm achieves performance superior to that of existing graph clustering algorithms in terms of three standard evaluation metrics: ACC, NMI, and ARI. Moreover, we use visualization results to analyze the effectiveness of the capsules and demonstrate that GCCN can learn cluster-oriented embeddings.

Sufficient real information in generator is a critical point for the generation ability of GAN. However, GAN and its variants suffer from lack of this point, resulting in brittle training processes. In this paper, we propose a novel variant of GAN, Posterior Promoted GAN (P2GAN), which promotes generator with the real information in the posterior distribution produced by discriminator. In our frame-work, different from other variants of GAN, the discriminator maps images to a multivariate Gaussian distribution and extracts real information. The generator employs the real information by AdaIN and a latent code regularizer. Besides, reparameterization trick and pretraining is applied to guarantee a stable training process in practice. The convergence of P2GAN is theoretically proved. Experimental results on typical high-dimensional multi-modal datasets demonstrate that P2GAN has achieved comparable results with the state-of-the-art variants of GAN on unsupervised image synthesis.

Existing deep graph clustering methods usually rely on neural language models to learn graph embeddings. However, these methods either ignore node feature information or fail to learn cluster-oriented graph embeddings. In this paper, we propose a novel deep graph clustering framework to tackle these two issues. First, we construct a feature transformation module to effectively integrate node feature information with graph topologies. Second, we introduce a graph embedding module and a self-supervised learning strategy to constrain graph embeddings by leveraging the graph similarity and the self-learning loss to group similar graphs together, thus encouraging the obtained graph embeddings to be cluster-oriented. Extensive experimental results on eight real-world graph datasets validate the superiority of the proposed method over existing ones.

Graph convolutional neural networks (GCN) have been the model of choice for graph representation learning, which is mainly due to the effective design of graph convolution that computes the representation of a node by aggregating those of its neighbors. However, existing GCN variants commonly use 1-D graph convolution that solely operates on the object link graph without exploring informative relational information among object attributes. This significantly limits their modeling capability and may lead to inferior performance on noisy and sparse real-world networks. In this paper, we explore 2-D graph convolution to jointly model object links and attribute relations for graph representation learning. Specifically, we propose a computationally efficient dimensionwise separable 2-D graph convolution (DSGC) for filtering node features. Theoretically, we show that DSGC can reduce intra-class variance of node features on both the object dimension and the attribute dimension to learn more effective representations. Empirically, we demonstrate that by modeling attribute relations, DSGC achieves significant performance gain over state-of-the-art methods for node classification and clustering on a variety of real-world networks. The source code for reproducing the experimental results is available at https://github.com/liqimai/DSGC.

Few-shot classification classifies unlabeled samples into correct classes when only few training samples are available for each class. Current researches focus on extracting better features and learning similarities between support data and query data based on the extracted features. However, the location of the main objects and the background in an image are interfering factors for few-shot classification, as they contain little useful information. In this paper, we propose a class-based core feature extraction network (CCFEN) which highlights common object of images in the same class and reduces the interference of background by using local feature descriptors and core feature extraction network to learn the common object. Experiments on two classical few-shot learning datasets show that our method achieves better results than state-of-the-art few-shot learning methods.

User intent classification plays a vital role in dialogue systems.Since user intent may frequently change over time in many realistic scenarios, unknown (new) intent detection has become an essential problem, where the study has just begun.This paper proposes Understanding user intent is crucial for developing conversational and dialogue systems.It is essential to accurately identify the intent behind a user utterance to better guide downstream decisions and policies.With the advent of conversational AI, dialogue systems are becoming central tools in many applications such as mobile apps, companion bots, virtual assistants and so on.Since user interests may change frequently over time, the AI agents may continuously see unknown (new) user intents.Manual annotation can hardly catch up with such rapid development, which motivates the problem * Equal contribution.

Graph-based methods have been demonstrated as one of the most effective approaches for semi-supervised learning, as they can exploit the connectivity patterns between labeled and unlabeled data samples to improve learning performance. However, existing graph-based methods either are limited in their ability to jointly model graph structures and data features, such as the classical label propagation methods, or require a considerable amount of labeled data for training and validation due to high model complexity, such as the recent neural-network-based methods. In this paper, we address label efficient semi-supervised learning from a graph filtering perspective. Specifically, we propose a graph filtering framework that injects graph similarity into data features by taking them as signals on the graph and applying a low-pass graph filter to extract useful data representations for classification, where label efficiency can be achieved by conveniently adjusting the strength of the graph filter. Interestingly, this framework unifies two seemingly very different methods -- label propagation and graph convolutional networks. Revisiting them under the graph filtering framework leads to new insights that improve their modeling capabilities and reduce model complexity. Experiments on various semi-supervised classification tasks on four citation networks and one knowledge graph and one semi-supervised regression task for zero-shot image recognition validate our findings and proposals.

Han Liu, Xiaotong Zhang, Lu Fan, Xuandi Fu, Qimai Li, Xiao-Ming Wu, Albert Y.S. Lam. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Attributed graph clustering is challenging as it requires joint modelling of graph structures and node attributes. Recent progress on graph convolutional networks has proved that graph convolution is effective in combining structural and content information, and several recent methods based on it have achieved promising clustering performance on some real attributed networks. However, there is limited understanding of how graph convolution affects clustering performance and how to properly use it to optimize performance for different graphs. Existing methods essentially use graph convolution of a fixed and low order that only takes into account neighbours within a few hops of each node, which underutilizes node relations and ignores the diversity of graphs. In this paper, we propose an adaptive graph convolution method for attributed graph clustering that exploits high-order graph convolution to capture global cluster structure and adaptively selects the appropriate order for different graphs. We establish the validity of our method by theoretical analysis and extensive experiments on benchmark datasets. Empirical results show that our method compares favourably with state-of-the-art methods.

Multi-task clustering improves the clustering performance of each task by transferring knowledge among the related tasks. An important aspect of multi-task clustering is to assess the task relatedness. However, to our knowledge, only two previous works have assessed the task relatedness, but they both have limitations. In this paper, we propose a multi-task clustering with model relation learning (MTCMRL) method, which automatically learns the model parameter relatedness between each pair of tasks. The objective function of MTCMRL consists of two parts: (1) within-task clustering: clustering each task by introducing linear regression model into symmetric nonnegative matrix factorization; (2) cross-task relatedness learning: updating the parameter of the linear regression model in each task by learning the model parameter relatedness between the clusters in each pair of tasks. We present an effective alternating algorithm to solve the non-convex optimization problem. Experimental results show the superiority of the proposed method over traditional single-task clustering methods and existing multi-task clustering methods.

Multi-task clustering improves the clustering performance of each task by transferring knowledge across related tasks. Most existing multi-task clustering methods are based on the ideal assumption that the tasks are completely related. However, in real applications, the tasks are usually partially related. In these cases, brute-force transfer may cause negative effect which degrades the clustering performance. In this paper, we propose two multi-task clustering methods for partially related tasks: the self-adapted multi-task clustering (SAMTC) method and the manifold regularized coding multi-task clustering (MRCMTC) method, which can automatically identify and transfer related instances among the tasks, thus avoiding negative transfer. Both SAMTC and MRCMTC construct the similarity matrix for each target task by exploiting useful information from the source tasks through related instances transfer, and adopt spectral clustering to get the final clustering results. But, they learn the related instances from the source tasks in different ways. Experimental results on real data sets show the superiorities of the proposed algorithms over traditional single-task clustering methods and existing multi-task clustering methods on both completely and partially related tasks.

Multi-task clustering and multi-view clustering have severally found wide applications and received much attention in recent years. Nevertheless, there are many clustering problems that involve both multi-task clustering and multi-view clustering, i.e., the tasks are closely related and each task can be analyzed from multiple views. In this paper, we introduce a multi-task multi-view clustering framework which integrates within-view-task clustering, multi-view relationship learning, and multi-task relationship learning. Under this framework, we propose two multi-task multi-view clustering algorithms, the bipartite graph based multi-task multi-view clustering algorithm, and the semi-nonnegative matrix tri-factorization based multi-task multi-view clustering algorithm. The former one can deal with the multi-task multi-view clustering of nonnegative data, the latter one is a general multi-task multi-view clustering method, i.e., it can deal with the data with negative feature values. Experimental results on publicly available data sets in web page mining and image mining show the superiority of the proposed multi-task multi-view clustering algorithms over either multi-task clustering algorithms or multi-view clustering algorithms for multi-task clustering of multi-view data.

Traditional clustering algorithms deal with a single clustering task on a single dataset. However, there are many related tasks in the real world, which motivates multitask clustering. Recently some multitask clustering algorithms have been proposed, and among them multitask Bregman clustering (MBC) is a very applicable method. MBC alternatively updates clusters and learns relationships between clusters of different tasks, and the two phases boost each other. However, the boosting does not always have positive effects on improving the clustering performance, it may also cause negative effects. Another issue of MBC is that it cannot deal with nonlinear separable data. In this article, we show that in MBC, the process of using cluster relationship to boost the cluster updating phase may cause negative effects, that is, cluster centroids may be skewed under some conditions. We propose a smart multitask Bregman clustering (S-MBC) algorithm which can identify the negative effects of the boosting and avoid the negative effects if they occur. We then propose a multitask kernel clustering (MKC) framework for nonlinear separable data by using a similar framework like MBC in the kernel space. We also propose a specific optimization method, which is quite different from that of MBC, to implement the MKC framework. Since MKC can also cause negative effects like MBC, we further extend the framework of MKC to a smart multitask kernel clustering (S-MKC) framework in a similar way that S-MBC is extended from MBC. We conduct experiments on 10 real world multitask clustering datasets to evaluate the performance of S-MBC and S-MKC. The results on clustering accuracy show that: (1) compared with the original MBC algorithm MBC, S-MBC and S-MKC perform much better; (2) compared with the convex discriminative multitask relationship clustering (DMTRC) algorithms DMTRC-L and DMTRC-R which also avoid negative transfer, S-MBC and S-MKC perform worse in the (ideal) case in which different tasks have the same cluster number and the empirical label marginal distribution in each task distributes evenly, but better or comparable in other (more general) cases. Moreover, S-MBC and S-MKC can work on the datasets in which different tasks have different number of clusters, violating the assumptions of DMTRC-L and DMTRC-R. The results on efficiency show that S-MBC and S-MKC consume more computational time than MBC and less computational time than DMTRC-L and DMTRC-R. Overall S-MBC and S-MKC are competitive compared with the state-of-the-art multitask clustering algorithms in synthetical terms of accuracy, efficiency and applicability.

Density-based techniques seem promising for handling datauncertainty in uncertain data clustering. Nevertheless, someissues have not been addressed well in existing algorithms. Inthis paper, we firstly propose a novel density-based uncertaindata clustering algorithm, which improves upon existing algorithmsfrom the following two aspects: (1) it employs anexact method to compute the probability that the distance betweentwo uncertain objects is less than or equal to a boundaryvalue, instead of the sampling-based method in previouswork; (2) it introduces new definitions of core object probabilityand direct reachability probability, thus reducing thecomplexity and avoiding sampling. We then further improvethe algorithm by using a novel assignment strategy to ensurethat every object will be assigned to the most appropriatecluster. Experimental results show the superiority of our proposedalgorithms over existing ones.

Multitask Bregman Clustering (MBC) alternatively updates clusters and learns relationship between clusters of different tasks, and the two phases boost each other. However, the boosting does not always have positive effect, it may also cause negative effect. Another issue of MBC is that it cannot deal with nonlinear separable data. In this paper, we show that MBC's process of using cluster relationship to boost the updating clusters phase may cause negative effect, i.e., cluster centroid may be skewed under some conditions. We propose a smart multi-task Bregman clustering (S-MBC) algorithm which identifies negative effect of the boosting and avoids the negative effect if it occurs. We then extend the framework of S-MBC to a smart multi-task kernel clustering (S-MKC) framework to deal with nonlinear separable data. We also propose a specific implementation of the framework which could be applied to any Mercer kernel. Experimental results confirm our analysis, and demonstrate the superiority of our proposed methods.