The scheduling of TV commercials is currently done manually, and its automation and efficiency remains a significant challenge for broadcasters. This study aims to automate this scheduling process. In this study, we modeled the scheduling task as four types of 0-1 integer programming problems and evaluated the characteristics of the obtained solutions. Among the proposed models, in particular, the weekly partitioning model and the greedy algorithm-based model are designed to use broadcast slots from the first half of the month and to use a certain percentage of all broadcast slots evenly, respectively. An evaluation experiment was conducted using a test dataset which is based on actual one-month scheduling data from a domestic broadcaster, and the results suggest that all of the proposed models can generally achieve the targeted GRP (Gross Rating Point) goals outlined in the contracts. Furthermore, our findings reveal that the weekly partitioning model and the greedy algorithm-based model can provide the solutions as intended by the design.

This study proposes a reinforcement learning method incorporating relative vector-based rules to improve deadlock avoidance and task efficiency in object transportation problems. The proposed method enables each agent to learn optimal actions independently without sharing rewards among agents. By generating relative vectors from current and past positions, agents can achieve accurate environmental perception and efficient learning even under partially observable conditions. Experimental results demonstrated that the proposed approach mitigates mutual interference among agents, promotes the acquisition of temporary stopping behaviors, and improves overall task performance, as measured by the number of delivered items. In addition, some agents exhibited altruistic behavior, such as yielding to others, despite the absence of any explicitly encoded cooperation mechanism. These behaviors emerged as a result of individual reward optimization during learning. The findings indicate that reinforcement learning without shared rewards can still lead to the autonomous emergence of cooperative behavior, offering a practical and efficient learning framework for dynamic and complex environments.

This paper designs and evaluates a low-latency Adaptive Neuro-Fuzzy Inference System (ANFIS) implemented on an FPGA for real-time control systems. With the recent advancement of AI-IoT, there is an increasing demand for low-latency and low-power AI inference on edge devices. We designed a trained ANFIS model using C language for CPU and Verilog HDL for FPGA, and compared their classification performance, processing speed, and power consumption using the Iris and Balance Scale datasets. For the Iris dataset, the FPGA implementation achieved approximately 1.7 times faster processing speed and significantly lower power consumption compared to the CPU. Furthermore, our experiments with the Balance Scale dataset revealed implementation challenges related to the precision limits of 16-bit floating-point arithmetic, which can lead to instability in inference. The results demonstrate that FPGA-based ANFIS is effective for applications requiring real-time control and edge AI.

For developing convolutional neural networks (CNNs) for medical applications to replace invasive examinations with non-invasive ones, collecting a training dataset requires performing invasive examinations on patients. Therefore, in situations where a CNN is developed with a limited-sized training dataset, pointing out that the dataset size is too small without evidence is not desirable. We should collect new samples only when an improvement in estimation performance can be expected. Therefore, we verified whether we should collect more samples for the medical CNN developed in the previous research. In particular, by using the dataset for developing a CNN for estimating pulmonary artery wedge pressure (PAWP) from a chest radiograph, we built a CNN while increasing the dataset size and observed the changes in estimation performance and saliency maps. As a result, we verified that the changes in estimation performance do not converge. Moreover, during estimation, the CNN developed with a small number of samples checks a wide cardiac region, while the one developed with a large number of samples checks a narrower cardiac region. From this, in this case, increasing the training data is expected to improve both generalization performance and saliency maps.

An automatic seating detection system is developed for an automatic hand nut runner that tightens self-locking nuts. Our research is of particular significance in the field of aircraft assembly, where the use of special self-locking nuts is common. This nut has a different shape from commonly used nuts and does not loosen easily. Therefore, irregular torque values are observed during tightening, making it challenging to detect seating automatically. This paper proposes a method for detecting the seating of self-locking nuts using evolutionary computation techniques. The proposed method is based on anomaly detection by Hotelling. First, sensor values are used to create an indicator, and when the indicator meets a criterion, seating is determined. The indexes used for the judgment are generated by evolutionary computation. Numerical experiments validate the proposed method and demonstrate its practical application.

In recent years, databases of acted-emotional speech with labeled acted-emotions have often been used as training data for machine learning tasks in Speech Emotion Recognition and Emotion Speech Synthesis, instead of spontaneous speech databases labeled with the speaker’s own emotions. However, there has been insufficient analysis regarding the extent to which acted emotions correspond with perceived emotions, as well as comparisons of acting methods for naturalistic emotional expression. To address this, we constructed the Hiroshima City University Emotion Speech Database (HCUDB) for comparing and analyzing acted-emotion labels versus perceived emotion labels, and empathetic acting versus technical acting in speech. HCUDB comprises two corpora: one is “Acted-Emotion vs Perceived Emotion Evaluation Corpus,” that contains emotion speeches with both acted-emotion labels and perceived emotion labels annotated (HCUDB1), and the other is “Empathetic vs Technical Acting Method Comparison Corpus,” that contains emotion speeches performed by the same speaker using different acting methods (HCUDB2). This paper describes the methods of voice recording and emotion labeling for these corpora and presents statistical analysis of the acoustic features of the speech data in HCUDB.

In security technology to prevent unauthorized access, continuous authentication is a method that performs authentication multiple times continuously, as well as just once initially. In this study, we propose an improved method for DPTM (Dynamic Probability Trust Model), which is one of the continuous authentication algorithms. Specifically, the equations related to the trust value calculation in the conventional DPTM are modified to prevent the immediate lockout of genuine users. Experiments using hardware keyboard keystroke data collected from 20 subjects demonstrate that the improved DPTM exhibits superior performance in continuous authentication tasks compared to the conventional DPTM.

Time series forecasting is one of the essential information technologies for efficientanomaly detection and decision-making. Since most time series data contain various non-stationary features that are difficult to eliminate completely, it is important to learn continuously in an online manner. The Cortical Learning Algorithm (CLA), which mimics the human neocortex, is suitable for online learning and time series prediction. In this study, to make CLA more adapted to online learning, we propose a method to handle inputs that exceed predefined upper and lower input bounds. Experimental results show that not only can the upper and lower input constraints be eliminated, but also the prediction accuracy can be improved by maintaining an appropriate level of input representation granularity for the learner.

A BP learning method has been proposed for secure distributed processing using decomposed data. Although this method is a secure and safe learning method, it has the drawbacks of a large number of parameters and a large number of communications between servers. To improve these drawbacks, we propose a learning method in which the decomposed parameters are updated only at the central server. We show the number of parameters and communication cost of the conventional and proposed methods, and evaluate their accuracy through numerical experiments. As a result, we show that the model with Q servers can achieve an accuracy comparable to the conventional model with 1/Q decomposition parameters.

We propose new observer-based fuzzy controllers for general Takagi-Sugeno fuzzy system with nonlinear output equations and unmeasurable premise variables. For Takagi-Sugeno fuzzy systems with the unmeasurable premise variables, the separation principle may not hold in general. To overcome this difficulty, we employ the differential mean value theorem and the sector nonlinearity approach to reformulate as an appropriate error system in which the errors between the actual states and its estimates follow. Then, with the state feedback controller and the error system, we have an augmented closed-loop system that can independently and simultaneously analyze the stability of the states and the errors. Since our designed conditions do not require the Lipschitz condition, our approach is more relaxed than the existing approach. Finally, an illustrative example is given to show the effectiveness of the proposed approach.