Transfer learning can leverage knowledge from source tasks to improve learning on a target task, even when training samples are limited. However, most previous transfer learning approaches focus on a single view of the data and assume no uncertainty in the training samples. To address these limitations, we propose a novel method called boosting one-class transfer learning for multi-view uncertain data (UMTO-SVMs), which handles one-class classification in multi-view data with uncertain information. Our method transfers knowledge containing uncertainty from multiple source tasks to the target task and constrains complementary information across different views to improve consistency. By combining basic classifiers using the Adaboost algorithm, we build a robust classifier. We also design an iterative framework to optimize the method and prove the convergence of the algorithm. Experimental results on three benchmark datasets show that UMTO-SVMs outperform previous one-class classification methods.

Multi-instance learning (MIL) is a generalized form of supervised learning that attempts to extract useful information from sets of instances, known as bags. In practice, besides positive and negative bags, we can also collect universum bags. These universum bags contain instances that do not fit within the defined positive or negative classes; they may belong to a third class or other categories, similar to universum instances in classical machine learning. Additionally, dictionary learning can be used to eliminate noise hidden in the data and enhance the performance of the learning tool. In this paper, we propose a new dictionary-based multi-instance learning method with universum data (UDMIL). In the proposed model, universum bags are considered prior knowledge within the training data for classifier construction. We construct three types of dictionaries for positive bags, negative bags, and universum bags to enhance the sparsity of the training data and develop a better classifier. In addition, we introduce an alternative learning framework to solve the proposed model and acquire the MIL classifier for prediction. Extensive experiments show that the proposed method achieves superior performance.

Federated learning (FL) belongs to distributed machine learning. It allows data information sharing between users while protecting their data privacy at the same time. However, in many real-world scenarios, the data collected by client devices may be affected by noise in the working environment, leading to the decreased accuracy and confidence. Therefore, it is necessary to take measures to reduce data uncertainty in order to enhance the performance of FL algorithms. Traditional FL methods encounter challenges in handling uncertain data, which motivates the introduction of multi-view learning in this paper which designs an FL model suitable for highly variable data characteristics. We first achieve information complementarity among data views while ensuring the consistency in data representation. Furthermore, we quantify data uncertainty using the reachable region of data noise, thereby improving model robustness. To maintain data privacy between clients, we design an adaptive Kalman filter-based differential protection security protocol. Clients use the protocol to process local data and upload it to the master server, which returns the updated model parameters to the clients. The experimental results demonstrate the effectiveness of the federated learning model proposed in this paper.

The Cloth-changing person re-identification (CC-ReID) is more challenging than person re-identification (ReID) because of the unreliability of cloth-relevant features. Existing CC-ReID methods output the unique latent vector feature for the same pedestrian image. However, on the one hand, the latent vector feature generated by the pooling layer will lose more spatial information. On the other hand, the features that they focus on are the same for the same pedestrian image under different comparison pedestrian images, which does not consider the pedestrian pair relation information. For this, we propose a Multitask Tensor-based Relation Network (MTTRN) for CC-ReID. In MTTRN, we utilize the augmented cloth-changing identity images, human parsing images and head images to guide the model learning more fine-grained cloth-irrelevant feature cues. We propose a novel channel subspace generator and use the tensor mode-n product to generate the subspace tensor features instead of the vector feature generated by the pooling layer, which can retain more spatial feature information. Furthermore, we assume that the model will focus on different features for the same pedestrian image under different comparison pedestrian images in the same feature subspace or the same pedestrian image pair in different feature subspaces, in which the tensor feature pair relation is considered fully to mine more robust features. Extensive experiments show that our method achieves the state-of-the-art or competitive performance on three CC-ReID benchmark datasets and demonstrate the robustness of our model.

Few-shot Learning (FSL) aims to recognize the novel classes from few novel samples. Recently, lots of methods have been proposed to improve FSL performance by introducing privileged information. However, on the one hand, they utilize the class name or class-level description generated by some tools such as WordNet as the privileged information. On the other hand, they are all one-generation based and just use the simple convex integration of visual modality and privileged information modality. Besides, the classic FSL dataset miniImageNet has no labels for few-shot instance-level privileged information. In this paper, we propose that the few-shot instance-level privileged information generated by few-shot visual images samples are more concrete and more diverse, which is more in line with the real world situation than the class-level privileged information that is the abstract concept summarized from a large number of visual image samples. For this, we propose a novel Two-generation based FSL method (2G-FSL) which transfers the prior knowledge from the prior model to the posterior model. This can make 2G-FSL learns the meta-knowledge about preserving correct prior knowledge and self-correcting erroneous prior knowledge after introducing the few-shot instance-level privileged information, growing into a more robust posterior model. In 2G-FSL, we introduce a novel Latent Feature Augmentation (LFA) module in posterior model to learn the episode-related augmentation and integration of the latent features of visual and privileged information modalities instead of the simple convex integration, which can generate diverse modality integration strategies for enhancing the diversity of latent features to make the features more robust to alleviate the insufficient data problem of FSL. We make the dataset of few-shot instance-level privileged information of miniImageNet publicly available for the subsequent research of FSL with few-shot instance-level privileged information. Experimental results demonstrate the effectiveness and superiority of 2G-FSL with LFA in FSL with few-shot instance-level privileged information.

Few-shot object counting (FSOC) estimates object quantities in query images using a few of support information. Unlike traditional counting methods, FSOC prioritizes more discriminative and generalized similarity measures between query and support data. This facilitates counting objects from new categories without extensive dataset creation or costly retraining. However, existing approaches often rely on fixed similarity rules, leading to spatial information loss. Limited training data can yield sparse similarity feature distribution, hampering the model's learning and its ability to handle objects with large intraclass differences. In this study, we introduce a novel FSOC network named DSALVANet that comprises the dynamic similarity-aware module (DSAM) and the latent variable augmentation module (LVAM). DSAM establishes adaptive metric rules for support features to find similar regions in the metric space for accurate object counting. LVAM utilizes prior similarity knowledge from DSAM to model the latent distribution of the density map, improving the decoder's robustness by sampling diverse latent variables during training. Extensive experiments on the FSOC benchmark and remote-sensing datasets demonstrate our method's effectiveness and state-of-the-art performance.