Label Distribution Learning (LDL) is a groundbreaking paradigm for addressing the task with label ambiguity. Subjectivity in annotating label description degrees often leads to imbalanced label distribution. Existing approaches either adopt representation alignment or decoupling strategies to solve the imbalanced label distribution learning (ILDL). However, representation alignment-based methods overlook the issue of gradient vanishing for non-dominant branches within imbalanced label distributions, while decoupling-based approaches fail to achieve adaptive weight optimization. To address these issues, we propose Adaptive Momentum and Exponential Moving Average weighted modeling (AMEMA). AMEMA combines EMA-based loss weighting with momentum allocation to mitigate gradient attenuation in non-dominant label learning and adaptively balance the optimization signals between dominant and non-dominant branches. It computes and updates Kullback-Leibler divergence losses for each branch using EMA, and applies different initial momenta to facilitate branch-specific optimization dynamics. Dynamic weighting coefficients, derived from EMA-smoothed losses, allow the model to adjust its learning direction adaptively and improve the learning of non-dominant labels. Extensive experiments on benchmark datasets show that AMEMA consistently outperforms state-of-the-art ILDL methods across various evaluation metrics.

The goal of fusing hyperspectral images (HSI) and multispectral images (MSI) is to generate high-resolution hyperspectral images for downstream tasks. However, most existing methods overlook the specific requirements of these tasks, leading to a gap between the fusion process and its subsequent applications due to insufficient guidance from downstream tasks. To address this issue, we propose a classification-guided multitask interactive network (CGNet) that integrates both fusion and classification tasks into a unified framework, with two branches producing the fused image and classification results, respectively. In the fusion branch, we design a multi-level residual refinement module to efficiently integrate spatial and spectral information. Additionally, an attention-based multi-scale fusion module, incorporating both spatial and channel attention, is carefully crafted to enhance representation learning. In the classification branch, both 2-D and 3-D convolutions are employed to improve classification performance. Moreover, an information interaction module is proposed to guide the fusion task based on classification outcomes. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches on the Pavia Centre and Pavia University datasets.

Hyperspectral super-resolution involves combining low-resolution hyperspectral images with high-resolution multispectral images to produce a high-resolution hyperspectral image. Recently, although many methods for hyperspectral image super-resolution have been proposed, they often fail to fully utilize the high similarity among adjacent bands to enhance fusion performance. Therefore, we propose a grouping strategy-based progressive fusion network (GPFNet) for hyperspectral super-resolution. The core of GPFNet is the grouping strategy fusion block (GPF block), in which grouping-based spatial-spectral information fusion and spatial information refinement are performed. We design the spatial-spectral information fusion module (SSIFM) based on grouped convolutions to capture the feature differences from adjacent bands. To refine spatial details, we develop the spatial information enhancement module (SpaEM), which leverages the hierarchical features extracted by the multi-scale feature extraction module (MIEM). Additionally, a progressive fusion strategy, which involves using multiple upsampled hyperspectral images and downsampled multispectral images, further preserves spectral integrity and spatial details. Extensive experiments show that GPFNet outperforms state-of-the-art methods both qualitatively and quantitatively.

Label Distribution Learning (LDL) has been successfully implemented in numerous practical applications. However, the imbalance in label distributions presents a significant challenge due to the substantial variation in annotation information. To tackle this issue, we introduce Decoupled Imbalance Label Distribution Learning (DILDL), which decomposes the imbalanced label distribution into a dominant label distribution and a non-dominant label distribution. Our empirical findings reveal that an excessively high description degree of dominant labels can result in substantial gradient information attenuation for non-dominant labels during the learning process. Therefore, we employ the decoupling approach to balance the description degrees of both dominant and non-dominant labels independently. Furthermore, we align the feature representations with the representations of dominant and non-dominant labels separately, aiming to effectively mitigate the distribution shift problem. Experimental results demonstrate that our proposed DILDL outperforms other state-of-the-art methods for imbalance label distribution learning.

Domain adaptation (DA)-based cross-domain hyperspectral image (HSI) classification methods have garnered significant attention. The majority of DA techniques utilize models based on convolutional neural networks (CNNs) and Transformers for feature extraction. However, Transformers may struggle to capture local details in HSIs, while CNNs often underperform in handling long-range dependencies. Furthermore, many methods focus only on aligning marginal distributions while ignoring the consistency of inter-class features, which may lead to feature confusion and degraded classification accuracy. To overcome the challenges mentioned, we propose a Mamba-CNN and knowledge distillation network (MKDnet). Firstly, the network employs a feature extractor that integrates Mamba and CNN frameworks for cross-domain HSI classification, enabling the capture of both global and local features while effectively capturing long-range dependencies. Secondly, domain alignment is achieved through distribution alignment and graph alignment. In the distribution alignment phase, we design a knowledge distillation architecture that utilizes soft labels to enhance the understanding of relationships between classes, thereby improving the consistency of inter-class features. In the graph alignment phase, we use graph convolution to capture connections between nodes and edges and transfer class-level topological relationships across domains. Finally, the classifier is used to obtain classification results, with consistency constraints applied to balance features between classes more effectively. Extensive experiments have demonstrated that MKDnet outperforms other state-of-the-art methods on three public cross-domain HSI datasets.

Reinforcement learning (RL) aims to formulate the recommendation task as a Markov decision process (MDP) and trains an agent to automatically learn the optimal recommendation policy from interaction trajectories through trial-and-error and reward mechanisms. However, most existing RL-based approaches overlook the correlation between items and the dynamics of user interests implied in temporally close interactions. Therefore, in this paper, we propose a reinforcement learning method that incorporates a “recent-k items” distribution to capture users' local preferences. Specifically, we model the output layer as two distinct branches. The “recent-k items” branch, formulated with a Kullback-Leibler divergence loss, learns the recent interests of users, whereas the other branch utilizes a one-step temporal difference error to capture long-term preferences. The proposed structure is integrated into deep Q-learning and actor-critics, resulting in two enhanced methods named R$k$Q and R$k$AC, respectively. Furthermore, a novel soft inter-reward is carefully designed to enhance the proposed method, and we theoretically prove the convergence of the proposed algorithm. We perform extensive experiments on two large real-world datasets and conduct further analysis of the influences of different action sequences, time intervals, and enhancement capabilities for state-of-the-art models. The experimental results demonstrate the efficacy of our proposed methods.

The purpose of infrared and visible image fusion is to combine the advantages of different types of images to produce more robust and informative images. However, if the source images are noisy, existing fusion methods may not produce clear results. To address this issue, we propose a novel method for infrared and visible image fusion with noise reduction. This method enhances the visual perception of fused images by integrating features of different scales extracted by the denoising network into the fusion network. By using deformable convolutional denoising networks, noise in images can be removed and features can be enhanced. Then, a set of reinforced attention fusion modules (RAFM) are designed to fuse the features extracted by the denoising network. Experimental results demonstrate the effectiveness of our proposed method, which outperforms existing state-of-the-art methods in terms of fusion accuracy and visual perception.

Label distribution learning (LDL) is a novel machine learning paradigm for solving ambiguous tasks, where the degree to which each label describing the instance is ambiguous. However, obtaining the label distribution is high cost and the description degree is difficult to quantify. Most existing research works focus on designing an objective function to obtain the whole description degrees at once but seldom care about the sequentiality in the process of recovering the label distribution. In this article, we formulate the label distribution recovering task as a sequential decision process called sequential label enhancement (Seq_LE), which is more consistent with the process of annotating the label distribution in human brains. Specifically, the discrete label and its description degree are serially mapped by the reinforcement learning (RL) agent. Besides, we carefully design a joint reward function to drive the agent to fully learn the optimal decision policy. Extensive experiments on 16 LDL datasets are conducted under various evaluation metrics. The experimental results demonstrate convincingly that the proposed sequential label enhancement (LE) leads to better performance over the state-of-the-art methods.

Reinforcement learning maps from perceived state representation to actions, which is adopted to solve the video summarization problem. The reward is crucial for deal with the video summarization task via reinforcement learning, since the reward signal defines the goal of video summarization. However, existing reward mechanism in reinforcement learning cannot handle the ambiguity which appears frequently in video summarization, i.e., the diverse consciousness by different people on the same video. To solve this problem, in this paper label distributions are mapped from the CNN and LSTM-based state representation to capture the subjectiveness of video summaries. The dual-reward is designed by measuring the similarity between user score distributions and the generated label distributions. Not only the average score but also the the variance of the subjective opinions are considered in summary generation. Experimental results on several benchmark datasets show that our proposed method outperforms other approaches under various settings.

Label distribution learning (LDL) is a novel machine learning paradigm that gives a description degree of each label to an instance. However, most of training datasets only contain simple logical labels rather than label distributions due to the difficulty of obtaining the label distributions directly. We propose to use the prior knowledge to recover the label distributions. The process of recovering the label distributions from the logical labels is called label enhancement. In this paper, we formulate the label enhancement as a dynamic decision process. Thus, the label distribution is adjusted by a series of actions conducted by a reinforcement learning agent according to sequential state representations. The target state is defined by the prior knowledge. Experimental results show that the proposed approach outperforms the state-of-the-art methods in both age estimation and image emotion recognition.

In this paper, we demonstrate a practical system for automatic weather-oriented clothing suggestion, given the weather information, the system can automatically recommend the most suitable clothing from the user s personal clothing album, or intelligently suggest the most pairing one with the user-specified reference clothing. This is an extremely challenging problem due to the large discrepancy factors that should be considered under different weather conditions. To approach this task, we use clothing attributes as a mid-level bridge to narrow the gap between low-level features and the high-level weather categories. We adopt a scoring function, which includes three terms, to model the relationship. To acquire an optimized model and verify our proposed method, we collect a large clothing Weather-to-Garment (WoG) dataset. Experiments on the WoG dataset demonstrate that our learned model are effective for both weather-oriented clothing recommendation and pairing.