In existing research on data streams, most problems are processed and studied based on single data streams. However, there exist multiple data streams and the reaching data may contain noise information which is considered uncertain in representation. In this paper, we propose the multi-task transfer learning for multiple data streams with uncertain data (MTUMDS), which can make use of the similarity of multiple data streams to carry out multi-task learning that can improve the classification ability of the data stream model. At the same time, transfer learning is applied for each single data stream, which transfers knowledge from the known classifiers of the previous time windows to the current target window classifier. This can settle the situation that the concept drift causes model fitness reduction. Then, in view of the noise and collection error hidden in the real data, boundary constraints are generated for each sample to build the SVM classifier to solve the uncertainty of the data. A large number of experiments in multiple data streams show that our approach has better performance and robustness than previous studies.

Multi-view learning can improve classification performance by combining information between different views. Due to the similarity in different views of the dataset, sometimes the features obtained are highly limited and redundant. At the same time, different views accumulate a large amount of noisy information, which will affect the classification performance of the model. To solve these problems, we embed privileged information in the model and introduce dictionary learning, and proposed a new dictionary-based multi-view learning method with privileged information (MVDL-PI). First, two sets of dictionaries (synthetic dictionary and analysis dictionary) and sparse representation matrices of different information domains are obtained for each view information and privilege information through dictionary learning. Then, we obtain consistency information from the regularization terms of the two different sets of synthetic dictionaries and construct a LUPI (Learning using privileged information) classifier by the sparse representation. In addition, we use alternating convex optimization and Lagrange multiplier methods to optimize the model and prove its convergence. In the experiment, we did a number of experiments comparing this method with similar recent methods. The experimental results show that the MVDL-PI method is superior to other methods in terms of stability and classification accuracy.

In partial label learning (PLL) problem, each training sample corresponds to a group of candidate labels, in which only one label is the ground-truth label (correct label). Almost all the existing PLL algorithms attempt to eliminate the ambiguity of the candidate label sets by treating all labels indiscriminately. However, this kind of approach lacks the consideration of the complexity between the labels and the instances in the training process. Encouraged by the extensive researches of the self-paced learning (SPL) and transfer learning (TL) in various fields, this paper introduces SPL and TL together to address the PLL problem and proposes a new SPL framework, which is called self-paced method for transfer partial label learning (SPTPLL). The proposed model utilizes transfer learning model to share the parameters and regularization terms of the Support Vector Machine (SVM), which can transfer knowledge from the source task to the target task. Additionally, we implement the self-paced learning scheme by choosing a suitable self-paced function to enhance the robustness of the proposed model. In the process of learning iteration, the priority of training examples with their candidate labels is ranked through self-paced learning to control the learning process. Finally, we demonstrate the superior performance of the proposed method through a large number of experiments compared with state-of-the-art baseline methods.

Anomaly detection aims to identify unusual behavior or discriminate abnormal samples by referring to the normal samples of data. Most exiting anomaly detection approaches train the model using only the normal data due to the scarcity of anomalies. However, the negative data or anomalies do occur in many practical applications. In this paper, we propose a novel anomaly detection method called AdaDL-SVDD for addressing uncertain data problem. In this method, both normal and anomalous samples are utilized to generate sparse representations with dictionary learning in the training phase. Meanwhile, we incorporate Support Vector Data Description (SVDD) into framework to construct a minimum hypersphere for anomaly detection over the test data. Additionally, the AdaBoost method is considered to construct a strong classifier via combining the weak classifiers. In the end, the experimental results demonstrate that the proposed AdaDL-SVDD method achieves superior performance over the UCI datasets with uncertainty and noise.

Graph widely exists in the research study, with complex structure and rich information, and it has always been a powerful mode to analyze difficult problems. Meanwhile, graph classification is an important application problem in many fields, it means that we use the structure of known graph datasets to classify the unknown graph. Most of the current work is to classify through positive and unlabeled learning (PU learning) and multi-view learning respectively. For training and testing data, it is believed that they all have the same feature representation, but in fact, we can gain auxiliary features called privileged information for the training data. In the past, privileged information is always ignored in PU graph learning. To enhance the efficiency of the classifier, a new method called SMPUP is presented, which first extracts the features of graph from different perspectives, and introduces privileged information into the PU learning method to construct the classifier model. Further, we have introduced a self-paced learning (SPL) framework to ensure the model can learn efficiently. A large number of experiments with graph and biological datasets show that the SMPUP method performs better than previous methods.

Graph widely exists in our lives, with complex structure and rich information, and it has always been a powerful analysis mode to solve difficult problems. Graph classification is an important application problem in many fields, it means that we use the structure of a known graph dataset to classify the whole graph. Most of the current work is to classify through positive and unlabeled graph learning. For test data and training data, it is believed that they all have the same feature representation, but in fact, we can gain auxiliary features called privileged information that training data has but test data does not have. To enhance the efficiency of the classifier, a method called SMPUP is presented, which divides the graph data into different perspectives, introduces privileged information, and uses the PU learning method to classify them. We have also introduced a self-paced learning framework to ensure that the model can learn more efficiently. A large number of experiments with graph and biological datasets show that the SMPUP method can perform better than previous methods.