Direct policy search is a promising reinforcement learning framework particularly for controlling continuous, high-dimensional systems. Peters et al. proposed reward-weighted regression (RWR) as a direct policy search. The RWR algorithm estimates the policy parameter based on the expectation-maximization (EM) algorithm and is therefore prone to overfitting. In this study, we focus on variational Bayesian inference to avoid overfitting and propose direct policy search reinforcement learning based on variational Bayesian inference (VBRL). The performance of the proposed VBRL is assessed in several experiments involving a mountain car and a ball batting task. These experiments demonstrate that VBRL yields a higher average return and outperforms the RWR.

This paper proposes a matrix-based collaborative filtering (CF) employing personal values (MCFPV). Introduction of various factors such as diversity and long-tailedness in addition to accuracy is a recent trend in the study of recommender systems. We think recommending acceptable items while satisfying users’ preference is important when considering other factors than accuracy. Also, interpretability is one of important characteristics recommender systems should have. To recommend acceptable items on the basis of an interpretable mechanism, this paper proposes a matrix-based recommendation method based on personal values-based modeling. Whereas existing CF based on matrix factorization methods are known to be more accurate than neighborhood-based CF, latent factors obtained by existing methods are difficult to interpret. On the other hand, user/item models of the propose method (MCFPV) is expected to be interpretable, because it represents the effect of each attribute items have on user’s decision making. Regarding a model relationship matrix that connects user and item models, this paper proposes two approaches: manual setting and machine learning approaches. Experimental results using 5 datasets generated from actual review sites show that the proposed methods recommend much unpopular items than the state-of-the art matrix factorization-based methods while keeping precision and recall.

A software-defined network (SDN) partitions a network into a control plane and data plane. Utilizing centralized control, an SDN can accurately control the routing of data flow. In the network, links have various costs, such as bandwidth, delay, and hops. However, it is difficult to obtain a multicost optimization path. If online rerouting can be realized under multiple cost, then network performance can be improved. This paper proposes a multicost rerouting algorithm for elephant flow, as the latter is the main factor affecting network traffic. By performing path trimming, the algorithm can obtain the approximate optimal solution of (1+e) in polynomial time. Simulation results show that the proposed algorithm yields good performance.

In the field of cluster analysis, rough set-based extensions of hard C-means (HCM; k-means) including rough C-means (RCM), rough set C-means (RSCM), and rough membership C-means (RMCM) are promising approaches for dealing with the certainty, possibility, uncertainty of belonging of object to clusters. Since C-means-type methods are strongly affected by noise, noise clustering approaches have been proposed. In noise clustering approaches, noise objects, which are far from any cluster center, are rejected for robust estimation. In this paper, we introduce noise rejection approaches for rough set-based C-means based on probabilistic memberships and propose noise RCM with membership normalization (NRCM-MN), noise RSCM with membership normalization (NRSCM-MN), and noise RMCM (NRMCM). In addition, visualization demonstration of the cluster boundaries on the two-dimensional plane of the proposed methods is carried out to confirm the characteristics of each method. Furthermore, the clustering performance is verified by numerical experiments using real-world datasets.

Automatic detection of the skyline plays an important role in several applications, such as visual geo-localization, flight control, port security, and mountain peak recognition. Existing skyline detection methods are mostly used under common weather conditions; however, they do not consider bad weather situations, such as rain, which limits their application in real scenes. In this paper, we propose a multi-stream-stage DenseNet to detect skyline automatically under different weather conditions. This model fully considers the adverse factors influencing the skyline and outputs a probability graph of the skyline. Finally, a dynamic programming algorithm is implemented to detect the skyline in images accurately. A comparison with the existing state-of-the-art methods proves that the proposed model shows a good performance under rainy or common weather conditions and exhibits the best detection precision for the public database.

In heterogeneous wireless networks, service providers typically employ multiple radio access technologies to satisfy the requirements of quality of service (QoS) and improve the system performance. However, many challenges remain when using modern cellular mobile communications radio access technologies (e.g., wireless local area network, long-term evolution, and fifth generation), such as inefficient allocation and management of wireless network resources in heterogeneous wireless networks (HWNs). This problem is caused by the sharing of available resources by several users, random distribution of wireless channels, scarcity of wireless spectral resources, and dynamic behavior of generated traffic. Previously, resource allocation schemes have been proposed for HWNs. However, these schemes focus on resource allocation and management, whereas traffic class is not considered. Hence, these existing schemes significantly increase the end-to-end delay and packet loss, resulting in poor user QoS and network throughput in HWNs. Therefore, this study attempts to solve the identified problem by designing an enhanced resource allocation (ERA) algorithm to address the inefficient allocation of available resources vs. QoS challenges. Computer simulation was performed to evaluate the performance of the proposed ERA algorithm by comparing it with a joint power bandwidth allocation algorithm and a dynamic bandwidth allocation algorithm. On average, the proposed ERA algorithm demonstrates a 98.2% bandwidth allocation, 0.75 s end-to-end delay, 1.1% packet loss, and 98.9% improved throughput performance at a time interval of 100 s.

Rough set theory is studied to manage uncertain and inconsistent information. Because Pawlak’s decision logic for rough sets is based on the classical two-valued logic, it is inconvenient for handling inconsistent information. We propose a bilattice logic as the deduction basis for the decision logic of rough sets to address inconsistent and ambiguous information. To enhance the decision logic to bilattice semantics, we introduce Variable Precision Rough Set (VPRS). As a deductive basis for bilattice decision logic, we define a consequence relation for Belnap’s four-valued semantics and provide a bilattice semantic tableau TB4 for a deduction system. We demonstrate the soundness and completeness of TB4 and enhance it with weak negation.

The study examines the period of the Great Depression and analyzes several measures taken by the US President Roosevelt’s government that allowed the country to get out of the crisis. An analysis and proof of the correctness of the measures chosen to exit from the crisis was conducted using econometric models and the use of statistics. Techniques for overcoming crises are relevant for all countries. This study adapts an innovative perspective in that it used the sequence of the Cobb–Douglas type function including different types of factors, and applied fuzzy regression methods.

An emotion recognition framework based on a two-channel convolutional neural network (CNN) is proposed to detect the affective state of humans through facial expressions. The framework consists of three parts, i.e., the frontal face detection module, the feature extraction module, and the classification module. The feature extraction module contains two channels: one is for raw face images and the other is for texture feature images. The local binary pattern (LBP) images are utilized for texture feature extraction to enrich facial features and improve the network performance. The attention mechanism is adopted in both CNN feature extraction channels to highlight the features that are related to facial expressions. Moreover, arcface loss function is integrated into the proposed network to increase the inter-class distance and decrease the inner-class distance of facial features. The experiments conducted on the two public databases, FER2013 and CK+, demonstrate that the proposed method outperforms the previous methods, with the accuracies of 72.56% and 94.24%, respectively. The improvement in emotion recognition accuracy makes our approach applicable to service robots.