Since Deep Convolutional Neural Networks (DCNNs) and Vision Transformer perform well in learning generalizable image priors from large-scale data, these models have been widely used in image denoising tasks. However, vanilla DCNNs and Transformer suffer from two problems. First, the vanilla DCNNs and Transformer only accumulate the output along the channel axis, ignoring the internal relationship among channels. This results in the severely inadequate color structure representation retrieved from color images. Secondly, the DCNNs or Transformer-based image denoising models usually have a large number of parameters, high computational complexity, and slow inference speed. To resolve these issues, this paper proposes a highly-efficient Quaternion Transformer (QFormer) for image denoising. Specifically, the proposed Quaternion Transformer Block (QTB) simplifies the typical Transformer from a multi-branch structure to an elaborately sequential structure mainly with quaternion transformations, to alternately capture both long-range dependencies and local contextual features with color structure information. Furthermore, the proposed QTB can also avoid considerable element-wise multiplications of computing the self-attention matrices. Thus, our QTB can significantly reduce the computational complexity and its sequential structure can further improve the practical inference speed. Comprehensive experiments demonstrate that the proposed QFormer produces state-of-the-art results in both denoising performance and efficiency. We hope that our work will encourage further research to explore the Quaternion Transformer architecture for image denoising tasks.
论文检索
输入标题、作者或关键词，从 1,048 篇学术成果中精准定位
Label distribution learning (LDL) is an effective method to predict the label description degree (a.k.a. label distribution) of a sample. However, annotating label distribution (LD) for training samples is extremely costly. So recent studies often first use label enhancement (LE) to generate the estimated label distribution from the logical label and then apply external LDL algorithms on the recovered label distribution to predict the label distribution for unseen samples. But this step-wise manner overlooks the possible connections between LE and LDL. Moreover, the existing LE approaches may assign some description degrees to invalid labels. To solve the above problems, we propose a novel method to learn an LDL model directly from the logical label, which unifies LE and LDL into a joint model, and avoids the drawbacks of the previous LE methods. We also give the generalization error bound of our method and theoretically prove that directly learning an LDL model from the logical labels is feasible. Extensive experiments on various datasets prove that the proposed approach can construct a reliable LDL model directly from the logical label, and produce more accurate label distribution than the state-of-the-art LE methods. The code and the supplementary file can be found in https://github.com/seutjw/DLDL.
Neural predictors are effective in boosting the time-consuming performance evaluation stage in neural architecture search (NAS), owing to their direct estimation of unseen architectures. Despite the effectiveness, training a powerful neural predictor with fewer annotated architectures remains a huge challenge. In this paper, we propose a context-aware neural predictor (CAP) which only needs a few annotated architectures for training based on the contextual information from the architectures. Specifically, the input architectures are encoded into graphs and the predictor infers the contextual structure around the nodes inside each graph. Then, enhanced by the proposed context-aware self-supervised task, the pre-trained predictor can obtain expressive and generalizable representations of architectures. Therefore, only a few annotated architectures are sufficient for training. Experimental results in different search spaces demonstrate the superior performance of CAP compared with state-of-the-art neural predictors. In particular, CAP can rank architectures precisely at the budget of only 172 annotated architectures in NAS-Bench-101. Moreover, CAP can help find promising architectures in both NAS-Bench-101 and DARTS search spaces on the CIFAR-10 dataset, serving as a useful navigator for NAS to explore the search space efficiently.
We propose a meta-learning method for modeling Hamiltonian dynamics from a limited number of data. Although Hamiltonian neural networks have been successfully used for modeling dynamics that obey the energy conservation law, they require many data to achieve high performance. The proposed method meta-learns our neural network-based model using datasets in various dynamical systems, such that our model can predict vector fields of unseen systems. In our model, a system representation is inferred from given small data using an encoder network. Then, the system-specific vector field is predicted by modeling the Hamiltonian using a Gaussian process (GP) with neural network-based mean and kernel functions that depend on the inferred system representation. This GP-based Hamiltonian allows us to analytically obtain predictions that are adapted to small data while imposing the constraint of the conservation law. The neural networks are shared across systems, which enables us to learn knowledge from multiple systems, and use it for unseen systems. In our experiments, we demonstrate that the proposed method outperforms existing methods for predicting dynamics from a small number of observations in target systems.
An Efficient Prototype-Based Clustering Approach for Edge Pruning in Graph Neural Networks to Battle Over-Smoothing
PDF ↗Topology augmentation is a popular strategy to address the issue of over-smoothing in graph neural networks (GNNs). To prevent potential distortion of node representations, an essential principle is to enhance the separability between embeddings of nodes from different classes while preserving smoothness among nodes of the same class. However, differentiating between inter-class and intra-class edges becomes arduous when class labels are unavailable or the graph is partially labeled. While clustering offers an alternative for identifying closely connected groups of nodes, traditional clustering methods face challenges when applied to GNNs in terms of accuracy, efficiency, adaptability, and scalability to diverse graphs. To address these limitations, we introduce ClusterDrop, which uses learnable prototypes for efficient clustering and incorporates supervised signals to enhance accuracy and adaptability across different graphs. Experiments on six datasets with varying graph structures demonstrate its effectiveness in alleviating over-smoothing and enhancing GNN performance.
Although multi-view unsupervised feature selection (MUFS) is an effective technology for reducing dimensionality in machine learning, existing methods cannot directly deal with incomplete multi-view data where some samples are missing in certain views. These methods should first apply predetermined values to impute missing data, then perform feature selection on the complete dataset. Separating imputation and feature selection processes fails to capitalize on the potential synergy where local structural information gleaned from feature selection could guide the imputation, thereby improving the feature selection performance in turn. Additionally, previous methods only focus on leveraging samples' local structure information, while ignoring the intrinsic locality of the feature space. To tackle these problems, a novel MUFS method, called UNified view Imputation and Feature selectIon lEaRning (UNIFIER), is proposed. UNIFIER explores the local structure of multi-view data by adaptively learning similarity-induced graphs from both the sample and feature spaces. Then, UNIFIER dynamically recovers the missing views, guided by the sample and feature similarity graphs during the feature selection procedure. Furthermore, the half-quadratic minimization technique is used to automatically weight different instances, alleviating the impact of outliers and unreliable restored data. Comprehensive experimental results demonstrate that UNIFIER outperforms other state-of-the-art methods.
In multi-dimensional classification (MDC), each instance is associated with multiple class variables characterizing the semantics of objects from different dimensions. To consider the dependencies among class variables and the specific characteristics contained in different semantic dimensions, a novel deep MDC approach named PIST is proposed to jointly deal with the two issues via learning pairwise dimension-specific features. Specifically, PIST conducts pairwise grouping to model the dependencies between each pair of class variables, which are more reliable with limited training samples. For extracting pairwise dimension-specific features, PIST weights the feature embedding with a feature importance vector, which is learned via utilizing a global loss measurement based on intra-class and inter-class covariance. Final prediction w.r.t. each dimension is determined by combining the joint probabilities related to this dimension. Comparative studies with eleven real-world MDC data sets clearly validate the effectiveness of the proposed approach.
Negative-Binomial Randomized Gamma Dynamical Systems for Heterogeneous Overdispersed Count Time Sequences
PDF ↗Modeling count-valued time sequences has been receiving growing interests because count time sequences naturally arise in physical and social domains. Poisson gamma dynamical systems (PGDSs) are newly-developed methods, which can well capture the expressive latent transition structure and bursty dynamics behind count sequences. In particular, PGDSs demonstrate superior performance in terms of data imputation and prediction, compared with canonical linear dynamical system (LDS) based methods. Despite these advantages, PGDS cannot capture the heterogeneous overdispersed behaviours of the underlying dynamic processes. To mitigate this defect, we propose a negative-binomial-randomized gamma Markov process, which not only significantly improves the predictive performance of the proposed dynamical system, but also facilitates the fast convergence of the inference algorithm. Moreover, we develop methods to estimate both factor-structured and graph-structured transition dynamics, which enable us to infer more explainable latent structure, compared with PGDSs. Finally, we demonstrate the explainable latent structure learned by the proposed method, and show its superior performance in imputing missing data and forecasting future observations, compared with the related models.
Time series forecasting (TSF) plays a pivotal role in many real-world applications. Recently, the utilization of Large Language Models (LLM) in TSF has demonstrated exceptional predictive performance, surpassing most task-specific forecasting models. The success of LLM-based forecasting methods underscores the importance of causal dependence modeling and pre-trained knowledge transfer. However, challenges persist in directly applying LLM to TSF, i.e., the unacceptable parameter scales for resource-intensive model optimization, and the significant gap of feature space between structural numerical time series and natural language. To this end, we propose LeRet, a Language-empowered Retentive network for TSF. Technically, inspired by the causal extraction in LLM, we propose a causal dependence learner, enhanced by a patch-level pre-training task, to capture sequential causal evolution. To minimize the gap between numeric and language, we initialize a language description protocol for time series and design a TS-related language knowledge extractor to learn from language description, avoiding training with large-scale parameters. Finally, we dedicatedly achieve a Language-TS Modality Integrator for the fusion of two types data, and enable language-empowered sequence forecasting. Extensive evaluations demonstrate the effectiveness of our LeRet, especially reveal superiority on few-shot, and zero-shot forecasting tasks. Code is available at https://github.com/hqh0728/LeRet.
Medical image segmentation is one of the most critical tasks in medical image analysis. However, the performance of existing methods is limited by the lack of high-quality labeled data due to the expensive data annotation. To alleviate this limitation, we propose a novel multi-modal learning method for medical image segmentation. In our method, medical text annotation is incorporated to compensate for the quality deficiency in image data. Moreover, previous multi-modal fusion methods ignore the commonalities and differences between different modalities. Ideally, the fused features should maximize valuable information while minimizing redundant information. To achieve this goal, we propose a multimodal feature distribution learning method. It is adopted to model the commonalities and differences between text and image. Since medical image segmentation needs to predict detailed segmentation boundaries, we also design a prompt encoder to achieve fine-grained segmentation. Experimental results on three datasets show that the proposed method obtains superior segmentation performance. Source codes will be available at https://github.com/GPIOX/Multimodal.git.
The high communication cost is a major challenge in the federated learning (FL) training process. Several methods have been proposed to reduce communication costs on the uplink channel, primarily sparsification-based methods, which have overlooked the impact of downlink channels. However, model accuracy and communication cost issues arise when applying them in practical FL applications, especially when the bandwidth is limited both on the uplink and downlink channels. In this paper, we propose a novel secure-FL-compatible hybrid gradient compression framework (HGC) that handles both uplink and downlink communication. Specifically, HGC identifies and exploits three types of redundancies in the FL training process. With proposed optimization methods based on compression ratio correction and dynamic momentum correction, HGC improves the trade-off between communication cost and model performance. The extensive theoretical and empirical analysis demonstrates the effectiveness of our framework in achieving a high compression ratio for both uplink and downlink communications with negligible loss of model accuracy, surpassing the state-of-the-art compression methods.
Feature Norm Regularized Federated Learning: Utilizing Data Disparities for Model Performance Gains
PDF ↗Federated learning (FL) is a machine learning paradigm that aggregates knowledge and utilizes computational power from multiple participants to train a global model. However, a commonplace challenge—non-independent and identically distributed (non-i.i.d.) data across participants—can lead to significant divergence in model updates, thus diminishing training efficacy. In this paper, we propose the Feature Norm Regularized Federated Learning (FNR-FL) algorithm to tackle the non-i.i.d challenge. FNR-FL incorporates class average feature norms into the loss function by a straightforward yet effective regularization strategy. The core idea of FNR-FL is to penalize the deviations in the update directions of local models caused by the non-i.i.d data. Theoretically, we provide convergence guarantees for FNR-FL when training under non-i.i.d scenarios. Practically, our comprehensive experimental evaluations demonstrate that FNR-FL significantly outperforms existing FL algorithms in terms of test accuracy, and maintains a competitive convergence rate with lower communication overhead and shorter duration. Compared to FedAvg, FNR-FL exhibits a 66.24% improvement in accuracy and an 11.40% reduction in training time, underscoring its enhanced effectiveness and efficiency. The code is available on GitHub at: https://github.com/LonelyMoonDesert/FNR-FL.
Existing continual learning literature relies heavily on a strong assumption that tasks arrive with a balanced data stream, which is often unrealistic in real-world applications. In this work, we explore task-imbalanced continual learning (TICL) scenarios where the distribution of task data is non-uniform across the whole learning process. We find that imbalanced tasks significantly challenge the capability of models to control the trade-off between stability and plasticity from the perspective of recent prompt-based continual learning methods. On top of the above finding, we propose Dynamically Anchored Prompting (DAP), a prompt-based method that only maintains a single general prompt to adapt to the shifts within a task stream dynamically. This general prompt is regularized in the prompt space with two specifically designed prompt anchors, called boosting anchor and stabilizing anchor, to balance stability and plasticity in TICL. Remarkably, DAP achieves this balance by only storing a prompt across the data stream, therefore offering a substantial advantage in rehearsal-free CL. Extensive experiments demonstrate that the proposed DAP results in 4.5% to 15% absolute improvements over state-of-the-art methods on benchmarks under task-imbalanced settings. Our code is available at https://github.com/chenxing6666/DAP.
Data augmentation plays a critical role in improving model performance across various domains, but it becomes challenging with graph data due to their complex and irregular structure. To address this issue, we propose EPIC (Edit Path Interpolation via learnable Cost), a novel interpolation-based method for augmenting graph datasets. To interpolate between two graphs lying in an irregular domain, EPIC leverages the concept of graph edit distance, constructing an edit path that represents the transformation process between two graphs via edit operations. Moreover, our method introduces a context-sensitive cost model that accounts for the importance of specific edit operations formulated through a learning framework. This allows for a more nuanced transformation process, where the edit distance is not merely count-based but reflects meaningful graph attributes. With randomly sampled graphs from the edit path, we enrich the training set to enhance the generalization capability of classification models. Experimental evaluations across several benchmark datasets demonstrate that our approach outperforms existing augmentation techniques in many tasks.
SAEIR: Sequentially Accumulated Entropy Intrinsic Reward for Cooperative Multi-Agent Reinforcement Learning with Sparse Reward
PDF ↗Multi-agent reinforcement learning (MARL) performs well for solving complex cooperative tasks when the scenarios have well-defined dense rewards. However, there are usually sparse reward settings in many real-world multi-agent systems, which makes it difficult for MARL algorithms to successfully learn an effective strategy. To tackle this problem, we propose a novel sequentially accumulated entropy intrinsic reward named SAEIR, which utilizes the entropy of multi-agent system as a bonus to accelerate learning. Specifically, the multi-scale hypergraph critic is proposed to obtain high-order system state representation, which also enhances the ability to effectively evaluate the action produced by the actor. Based on the comprehensive and compact system state representation, the orderliness of multi-agent systems can be measured to determine the highly valuable states for adding entropy-based intrinsic rewards which leads to a highly efficient learning process. Empirical results demonstrate that our proposed method achieves state-of-the-art performance in several complex cooperative multi-agent environments with sparse reward settings.
Block-wise missing data poses significant challenges in real-world data imputation tasks. Compared to scattered missing data, block-wise gaps exacerbate adverse effects on subsequent analytic and machine learning tasks, as the lack of local neighboring elements significantly reduces the interpolation capability and predictive power. However, this issue has not received adequate attention. Most SOTA matrix completion methods appeared less effective, primarily due to overreliance on neighboring elements for predictions. We systematically analyze the issue and propose a novel matrix completion method "BlockEcho" for a more comprehensive solution. This method creatively integrates Matrix Factorization (MF) within Generative Adversarial Networks (GAN) to explicitly retain long-distance inter-element relationships in the original matrix. Besides, we incorporate an additional discriminator for GAN, comparing the generator's intermediate progress with pre-trained MF results to constrain high-order feature distributions. Subsequently, we evaluate BlockEcho on public datasets across three domains. Results demonstrate superior performance over both traditional and SOTA methods when imputing block-wise missing data, especially at higher missing rates. The advantage also holds for scattered missing data at high missing rates. We also contribute on the analyses in providing theoretical justification on the optimality and convergence of fusing MF and GAN for missing block data.
Semi-supervised image classification, leveraging pseudo supervision and consistency regularization, has demonstrated remarkable success. However, the ongoing challenge lies in fully exploiting the potential of unlabeled data. To address this, we employ information entropy neural estimation to utilize the potential of unlabeled samples. Inspired by contrastive learning, the entropy is estimated by maximizing a lower bound on mutual information across different augmented views. Moreover, we theoretically analyze that the information entropy of the posterior of an image classifier is approximated by maximizing the likelihood function of the softmax predictions. Guided by these insights, we optimize our model from both perspectives to ensure that the predicted probability distribution closely aligns with the ground-truth distribution. Given the theoretical connection to information entropy, we name our method InfoMatch. Through extensive experiments, we show its superior performance. The source code is available at https://github.com/kunzhan/InfoMatch.
Suppose we are given an integer k and n boxes, labeled 1,2,…,n by an adversary, each containing a single number chosen from an unknown distribution; the n distributions not necessarily identical. We have to choose an order to sequentially open the boxes, and each time we open the next box in this order, we learn the number inside. If we reject a number in a box, the box cannot be recalled. Our goal is to accept k of these numbers, without necessarily opening all boxes, such that the accepted numbers are the k largest numbers in the boxes (thus their sum is maximized). This problem, sometimes called a free order multiple-choice secretary problem, is one of the classic examples of online decision making problems. A natural approach to solve such problems is to sample elements in random order; however, as indicated in several sources, e.g., Turan et al. NIST 2015 [35], Bierhorst et al. Nature 2018 [10], pure randomness is hard to get in reality. Thus, pseudorandomness has to be used, with a small entropy. We show that with a very small O(log log n) entropy an almost-optimal approximation of the value of k largest numbers can be selected, with only a polynomially small additive error, for k 1.
Sample Quality Heterogeneity-aware Federated Causal Discovery through Adaptive Variable Space Selection
PDF ↗Federated causal discovery (FCD) aims to uncover causal relationships among variables from decentralized data across multiple clients, while preserving data privacy. In practice, the sample quality of each client's local data may vary across different variable spaces, referred to as sample quality heterogeneity. Thus, data from different clients might be suitable for learning different causal relationships among variables. Model aggregated under existing FCD methods requires the entire model parameters from each client, thereby being unable to handle the sample quality heterogeneity issue. In this paper, we propose the Federated Adaptive Causal Discovery (FedACD) method to bridge this gap. During federated model aggregation, it adaptively selects the causal relationships learned under the "good" variable space (i.e., one with high-quality samples) from each client, while masking those learned under the "bad" variable space (i.e., one with low-quality samples). This way, each client only needs to send the optimal learning results to the server, achieving accurate FCD. Extensive experiments on various types of datasets demonstrate significant advantages of FedACD over existing methods. The source code is available at https://github.com/Xianjie-Guo/FedACD.
ParsNets: A Parsimonious Composition of Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning
PDF ↗This paper provides a novel parsimonious yet efficient design for zero-shot learning (ZSL), dubbed ParsNets, in which we are interested in learning a composition of on-device friendly linear networks, each with orthogonality and low-rankness properties, to achieve equivalent or better performance against deep models. Concretely, we first refactor the core module of ZSL, i.e., the visual-semantics mapping function, into several base linear networks that correspond to diverse components of the semantic space, wherein the complex nonlinearity can be collapsed into simple local linearities. Then, to facilitate the generalization of local linearities, we construct a maximal margin geometry on the learned features by enforcing low-rank constraints on intra-class samples and high-rank constraints on inter-class samples, resulting in orthogonal subspaces for different classes. To enhance the model's adaptability and counterbalance the over-/under-fittings, a set of sample-wise indicators is employed to select a sparse subset from these base linear networks to form a composite semantic predictor for each sample. Notably, maximal margin geometry can guarantee the diversity of features and, meanwhile, local linearities guarantee efficiency. Thus, our ParsNets can generalize better to unseen classes and can be deployed flexibly on resource-constrained devices.