Domain adaptation has been a fundamental technology for transferring knowledge from a source domain to a target domain. The key issue of domain adaptation is how to reduce the distribution discrepancy between two domains in a proper way such that they can be treated indifferently for learning. In this paper, we propose a novel domain adaptation approach, which can thoroughly explore the data distribution structure of target domain. Specifically, we regard the samples within the same cluster in target domain as a whole rather than individuals and assigns pseudo-labels to the target cluster by class centroid matching. Besides, to exploit the manifold structure information of target data more thoroughly, we further introduce a local manifold self-learning strategy into our proposal to adaptively capture the inherent local connectivity of target samples. An efficient iterative optimization algorithm is designed to solve the objective function of our proposal with theoretical convergence guarantee. In addition to unsupervised domain adaptation, we further extend our method to the semi-supervised scenario including both homogeneous and heterogeneous settings in a direct but elegant way. Extensive experiments on seven benchmark datasets validate the significant superiority of our proposal in both unsupervised and semi-supervised manners.
With the dramatic increase of dimensions in the data representation, extracting latent low-dimensional features becomes of the utmost importance for efficient classification. Aiming at the problems of weakly discriminating marginal representation and difficulty in revealing the data manifold structure in most of the existing linear discriminant methods, we propose a more powerful discriminant feature extraction framework, namely, joint sparse locality-aware regression (JSLAR). In our model, we formulate a new strategy induced by the nonsquared $L_{2}$ norm for enhancing the local intraclass compactness of the data manifold, which can achieve the joint learning of the locality-aware graph structure and the desirable projection matrix. Besides, we formulate a weighted retargeted regression to perform the marginal representation learning adaptively instead of using the general average interclass margin. To alleviate the disturbance of outliers and prevent overfitting, we measure the regression term and locality-aware term together with the regularization term by forcing the row sparsity with the joint $L_{2,1}$ norms. Then, we derive an effective iterative algorithm for solving the proposed model. The experimental results over a range of benchmark databases demonstrate that the proposed JSLAR outperforms some state-of-the-art approaches.
Geometric iteration (GI) is one of the most efficient curve- or surface-fitting techniques in recent years, which is famous for its remarkable geometric significance. In essence, GI can be thought of as the sum of iterative methods for solving systems of linear equations, such as progressive iterative approximation (PIA) which relies on the theory of Richardson iteration. Thus, when the curve- or surface-fitting error is at a desired level, we want to have as few iterations as possible to improve efficiency when dealing with large data sets. Based on the idea of successive over-relaxation (SOR) iteration, we formulate a faster PIA curve and surface interpolation scheme using classical non-uniform cubic B-splines, named SOR-PIA. The genetic algorithm is utilized to estimate the best approximate relaxation factor of SOR-PIA. Similar to standard PIA, SOR-PIA can also be regarded as a process in which the control points move in one direction, but it can greatly reduce the number of iterations in the iterative process with the same fitting accuracy. By comparing with the standard PIA and WPIA algorithms, the effectiveness of the SOR-PIA iterative interpolation algorithm can be verified.
Deep reinforcement learning is a data-driven method, which is very promising for alleviating traffic congestion through intelligent control of traffic lights. In this paper, the traffic signal of an intersection is divided into four independent phases and then controlled by deep Q-network (DQN) models respectively. Models can receive observations from their own angle of view, i.e., north-south straight, north-south left turn, east-west straight, east-west left turn, instead of extracting features from the whole scene. We suppose that it is beneficial for learning better policy if agents could sense the environment more precisely. DQN models are jointly trained under the revised QMIX framework to promote coordination capability. For decentralized execution, traffic lights of the phase with the highest Q-value will turn green. The experiments are done under SUMO, the results demonstrate that our method obtains higher reward and lower delay compared to controlling the holistic cycle by using a single DQN model.