Search

Classification models categorize objects into given classes, guided by training samples with input features and labels. In practice, however, labels can be corrupted by human error or mistakes, known as label noise, which degrades classification accuracy. To address this issue, recently, various works propose the algorithms to clean datasets with label noise. We categorize the algorithms in granular ways, and review the algorithms, such as sample selection, label correction, and select-and-correct algorithms, based on the categorization. In addition, we provide future research directions for cleaning datasets, considering practical challenges, such as class imbalance, class incremental learning, and corrupted input features.

The ability to detect and respond to vineyard spatial variation can lead to improved management—a practice known as precision viticulture. The goal of this study was to determine if remote sensors can enhance precision viticulture applications by detecting vineyard spatial variation. The hypothesis was that differences in vine spectral reflectance, as detected by remote sensors, would be associated with variations in viticultural variables due to known relationships with vine size, structure, and pigmentation. Riesling grapevines were geolocated within six commercial vineyards across Niagara, Ontario. Water status, vine size, winter hardiness, virus titer, yield components, and berry composition were measured on these vines. Remote sensing technologies subsequently collected multispectral data by unmanned aerial vehicles and by proximal sensing technology (GreenSeeker™), which were transformed into the Normalized Difference Vegetation Index (NDVI). Direct relationships between NDVI and vine size, water status, yield, berry weight, and titratable acidity were observed, as well as inverse relationships between NDVI and Brix and potentially volatile terpenes. Remote sensing demonstrated the ability to detect vineyard areas differing in measures of vine health, yield, and berry composition in certain sites and years; however, more research is needed to determine when these technologies should be used for precision viticulture applications.

As engine monitoring data has become more complex with an increasing number of sensors, fault prediction based on artificial intelligence (AI) has emerged. Existing fault prediction models using AI significantly improve the accuracy of predictions by effectively handling such complex data, but at the same time, the problem arises that the AI-based models cannot explain the rationale of their predictions to users. To address this issue, we propose a time-series explanatory fault prediction framework to provide an explainability even when using AI-based fault prediction models. It consists of a data feature reduction process, a fault prediction model training process using long short-term memory, and an interpretation process of the fault prediction model via an explainable AI method. In particular, the proposed framework can explain a fault prediction based on time-series data. Therefore, it indicates which part of the data was significant for the fault prediction not only in terms of sensor type but also in terms of time. Through extensive experiments, we evaluate the proposed framework using various fault data by comparing the prediction performance of fault prediction and by assessing how well the main pre-symptoms of the fault are extracted when predicting a fault.

Recently, studies on early-exit mechanisms have emerged to reduce the computational cost during the inference process of deep learning models. However, most existing early-exit architectures simply determine early exiting based only on a target confidence level in the prediction, without any consideration of the computational cost. Such an early-exit criterion fails to balance accuracy and cost, making it difficult to use in various environments. To address this problem, we propose a novel, cost-effective early-exit architecture in which an early-exit criterion is designed based on the Markov decision process (MDP). Since the early-exit decisions within an early-exit model are sequential, we model them as an MDP problem to maximize accuracy as much as possible while minimizing the computational cost. Then, we develop a cost-effective early-exit algorithm using reinforcement learning that solves the MDP problem. For each input sample, the algorithm dynamically makes early-exit decisions considering the relative importance of accuracy and computational cost in a given environment, thereby balancing the trade-off between accuracy and cost regardless of the environment. Consequently, it can be used in various environments, even in a resource-constrained environment. Through extensive experiments, we demonstrate that our proposed architecture can effectively balance the trade-off in different environments, while the existing architectures fail to do so since they focus only on reducing their cost while preventing the degradation of accuracy.

Purpose: This study aimed to develop hyaluronic acid (HA)-based, retinoic acid (RA)-containing nanomicelles and to investigate the effects of these newly developed nanomicelles on regeneration of the vaginal epithelium and aquaporin 3 (AQP3) expression in a murine menopause model. Materials and Methods: The HA-based, RA-loaded nanomicelles were developed, and the RA-loading rate, encapsulation efficiency, and hydrodynamic diameter were measured. Female BALB/c mice (8 weeks; n=30) were divided into control and experimental groups. Menopause was established in the experimental group by removing both ovaries. The experimental group was further divided into an ovariectomy group, an HA-C18 vehicle group, and an HA-C18-RA group (2.5 µg per mouse); vaginal administration of HA-C18 or HA-C18-RA was performed once daily. After 4 weeks of treatment, murine vaginal tissue was removed, and histological analysis was performed. Results: Three drug-loaded nanomicelles were synthesized: the RA content in HA-C18-RA-10, HA-C18-RA-20, and HA-C18-RA-30 was 3.13%, 2.52%, and 16.67%, respectively, and the RA encapsulation efficiency was 95.57%, 83.92%, and 93.24%, respectively. In the experimental versus control group, serum estrogen levels were significantly reduced, and the vaginal mucosal epithelial layer was significantly thinner. After 4 weeks of treatment, the thickness of the vaginal mucosal epithelial layer and AQP3 expression was increased in the HA-C18-RA group compared with the HA-C18 vehicle group. Conclusions: The newly developed HA-based nanomicelles containing RA resulted in vaginal epithelial recovery and increased AQP3 expression. The results may contribute to the development of functional vaginal lubricants or moisturizers for the treatment of vaginal dryness.

Deep reinforcement learning can effectively address resource allocation in wireless networks. However, its learning speed may be slower in more complex networks and a new policy should be learned for a newly-arrived system due to a lack of network adaptability. To address these issues, we propose a federated learning framework for resource allocation in wireless networks with multiple systems. It accelerates the learning speed by aggregating the policy at each system into a central policy and ensures network adaptability by using the central policy. Through experiments, we demonstrate that our proposed framework achieves both learning acceleration and network adaptability.

Ensuring operational reliability in machinery requires accurate fault detection. While time-domain vibration pulsation signals are intuitive for pattern recognition and feature extraction, downsampling can reduce analytical complexity, but may result in low-precision data, affecting fault detection performance. To address this, we propose time–frequency feature fusion, combining information from both the time and frequency domains for fault detection. Our approach transforms vibrational pulse data into instantaneous revolutions per minute (RPM) and employs statistical analysis for the time-domain features. For the frequency-domain features, we use the combined method of empirical mode decomposition and independent component analysis (EMD-ICA), along with the Wigner bispectrum method to capture the nonlinear characteristics and phase conjugation. Using a deep neural network (DNN), we classify the anomaly states, demonstrating the effectiveness and versatility of our approach in detecting anomalies and improving diagnostic precision. Compared to using time or frequency features alone, our time–frequency feature fusion model achieves higher accuracy, with 100% accuracy at lower downsampling rates and 96.3% accuracy at a downsampling rate of 100×.

Individual vineyards can vary spatially for several viticultural attributes, including water stress, nutrient status, growth/vigour and disease—which can, in turn, impact berry composition and resulting wine products. The goal of this study was to determine if vineyard variability detected by remote sensing using an unmanned aerial vehicle (UAV) could be used to zonally harvest vineyard blocks and produce wines that are sensorially differentiable. The specific hypothesis was that remote sensing would detect vineyard variation in viticultural variables and associate this variation with differences in wine sensory attributes based upon zonal harvesting. In six commercial Riesling vineyards across the Niagara Peninsula in Ontario, Canada, a UAV collected multispectral data, which were used to calculate the normalized difference vegetation index (NDVI). Grapevines (≈ 80) in a grid pattern were geo-located within each block and vineyard UAV NDVI maps were used for zonal harvesting of geo-located vines in areas corresponding to high vs. low NDVI. Wines made from these zones were then compared chemically and sensorially. Overall, wines created from high vs. low NDVI zones differed inconsistently in their basic wine composition. Sensorially, for certain sites and vintages, panellists distinguished between wines made from high vs. low NDVI zones using a sorting task. UAV NDVI demonstrated the ability to determine areas within a vineyard block that could produce wines that were sensorially distinguishable from one another.

Purpose: The expression of epithelial progenitor cells (EPCs) in rat vagina was recently reported. The aims were to investigate the effects of estrogen on vaginal EPCs in the oophorectomized female rat model. Materials and Methods: Female Sprague-Dawley rats (230–240 g, n=30) were divided into 3 groups: control (n=10), bilateral oophorectomy (OVX, n=10), and bilateral OVX followed by subcutaneous injections of 17β-estradiol (50 µg/kg/day, n=10). After 4 weeks, the expression of EPC-specific markers (CD44, estrogen receptor alpha [ERα], and progesterone receptor) were evaluated by immunohistochemistry and Western blot. Results: The CD44/ERα double-labeled cells were mainly expressed in basal cell layers and suprabasal layers as shown by confocal immunofluorescence. Confocal microscopy revealed that the number of CD44+/ERα+ cells decreased in the OVX group compared with the controls but was similar to control levels in rats receiving estrogen replacements. The protein expression of CD44 and ERα decreased after OVX and was restored to control levels after estrogen supplementation. Conclusions: Markers of EPCs were expressed in the vagina, and the expression of resident EPCs was regulated by estrogen. These findings imply that resident EPCs may have an important role in the regeneration of vaginal mucosa by estrogen replacement.

In this article, we study wireless power transfer (WPT) beam scheduling for a system which consists of IoT devices and a power beacon (PB) using switched beamforming. In such a system, the IoT devices have a non-static behavior (e.g., their location and power requests keep changing) in general, which conventional WPT beam scheduling algorithms are not capable of adaptively dealing with. To address the non-static behavior, we propose a procedure of deep neural network (DNN)-based WPT beam scheduling. In the procedure, the power-deficient IoT devices transmit a common pilot signal simultaneously. Then, the PB effectively provides power to them with a DNN-based WPT beam scheduling policy. In the DNN-based policy, an estimation of the non-static behavior from the received pilot signals and an adaptive beam generation considering the estimated non-static behavior are integrated thanks to the powerful representational capability of DNNs. To allow the DNN-based policy to learn the optimal policy, we propose a Deep WPT Beam scheduling policy Gradient (DWBG) algorithm using deep reinforcement learning. Through the simulation, we show that DWBG achieves a close performance to the optimal policy. This demonstrates that our algorithm can be applied for practical WPT IoT systems with non-static IoT devices.

This study focuses on revisiting the tectostratigraphic framework of the Ulleung Basin and conceptualizing neotectonics around the western East Sea margin. Based on the analysis of 2D and 3D multi-channel seismic reflection data and offshore drill wells, we divided the entire sedimentary successions of the Ulleung Basin into four tectostratigraphic sequences, named TS1 (c. 23–16 Ma), TS2 (c. 16–9 Ma), TS3 (c. 9–4 Ma), and TS4 (c. 4 Ma–present), in ascending order. The results show that each sequence has been deformed once or multiple times in different periods by juxtaposing two major compressional structures named the Dolgorae Thrust-Fold Belt and the Gorae Anticline. Interpretation of the stratal deformation and termination patterns of the syn- and post-deformational sequences of each structures suggests that the thrusting and folding of the Dolgorae Thrust-Fold Belt was active from c. 16 Ma to c. 9 Ma under the NNW–SSE compressional stress regime (Stage-2), whereas the Gorae Anticline was active from 4 Ma to the present under the ENE–WSW compressional stress regime (Stage-4). Between these two compressional events, there was an intervening period of regional slow subsidence driven by thermal contraction of the back-arc lithosphere and isostatic sedimentary loading (Stage-3). Based on the stratigraphic and structural reconstruction, we propose a 4-stage tectonic model: Stage-1) back-arc opening stage associated with the southward drift of the Japanese islands (c. 23–16 Ma), Stage-2) tectonic-inversion stage in association with the reorganization of the Pacific and Philippine Sea plates and clockwise rotation of SW Japan (c. 16–9 Ma), Stage-3) post-inversion stage with regional thermal and isostatic subsidence (c. 9–4 Ma), and Stage-4) neotectonic stage in which embryonic subduction is nucleating on the East Sea margins under the E–W compressional stress regime (c. 4 Ma–present).

In this paper, we consider a heterogeneous mobile cloud computing (HMCC) system that consists of remote cloud servers, local cloudlets, task offloading mobile devices (TMDs), non-task offloading MDs (NTMDs), and radio access networks such as cellular networks and WLANs. TMDs have the capability of task offloading to remote cloud servers or cloudlets, whereas NTMDs are conventional cellular users that do not have such capability. By using stochastic geometry, we analyze the outage probability of task offloading in the MCC system with only remote cloud servers and that in the HMCC with both remote cloud servers and cloudlets. The analysis provides useful information, i.e., how the varying system parameters affect the outage probability. From the analysis, we show that there is an intrinsic limitation in reducing the outage probability in the MCC system due to the outage when accessing remote cloud servers. In addition, we show that the use of cloudlets is a promising solution to overcome this limitation. However, a tradeoff exists in using cloudlets due to their deployment and operation costs. Thus, to address this tradeoff, we also study the optimal cloudlet deployment to maximize the cloud service provider's profit while guaranteeing maximum outage probability requirements.

Purpose: To investigate the feasibility of a polycaprolactone (PCL) scaffold fabricated by three-dimensional (3D) printing for tissue engineering applications for tunica albuginea. Materials and Methods: PCL scaffolds were fabricated by use of a 3D printing system. Two scaffolds were fabricated that differed in the architecture of the lay-down pattern: a 90°PCL scaffold and a 45°PCL scaffold. Mechanical properties were measured to compare tensile strength between the two scaffold types. The scaffolds were characterized by scanning electron microscope (SEM) images. The scaffolds were seeded with fibroblast cells, and the ability of these scaffolds to support the cells was evaluated by immunofluorescence staining. Results: The PCL scaffolds had well-structured shapes, regular arrays, and good interconnection in SEM images. The horizontal and vertical Young’s modulus coefficients were 13 and 12 MPa for the 90°PCL scaffold and 19 and 21 MPa for the 45°PCL scaffold, respectively. Microscopy images revealed that human fibroblast cells covered the entire scaffold surface. Immunofluorescence staining of ER-TR7 confirmed that the fibroblast cells remained viable and proliferated throughout the time course of the culture. Conclusions: This preliminary study provides experimental evidence for the feasibility of 3D printing of PCL scaffolds for tissue engineering applications of tunica albuginea.

Early detection and repeated learning training of children with intellectual disabilities are important factors that are directly related to the quality of future life of such children. However, implementation of such programs is challenging, as the subjects are still children. Therefore, motivation and interest are important for children with intellectual disabilities to carry out continuous training. In this study, Neuro-World, a game-based cognitive training program, was proposed for motivation and interest induction, and its effectiveness was compared with that of a conventional cognitive training program through analysis after both programs were implemented by professional therapists. The pre-test and post-test results of the game-based cognitive training program were statistically significant and showed superiority in the comparison with the conventional program. Therefore, the game-based cognitive training program developed in this study through digital media is expected to be effective in improving cognitive learning ability.

An optical probe was developed to measure the change of oxy-hemoglobin (OHb), deoxy- hemoglobin (RHb), and total hemoglobin (THb) along with temperature from the vaginal wall of female rats. Apomorphine (APO, 80 μg/kg) was administered to elicit sexual arousal in female Sprague Dawley rats (SD, 180−200 g). The behavior changes caused by APO administration were checked before monitoring vaginal responses. The changes of oxy-, deoxy-, and total hemoglobin concentration and the temperature from the vaginal wall were monitored before, during, and after APO administration. Animals were under anesthesia during the measurement. After APO administration, the concentration of OHb (55 ± 29 μM/DPF), RHb (33 ± 25 μM/DPF), and THb (83 ± 59 μM/DPF) in the vaginal wall increased in a few min, while saline administration did not cause any significant change. In case of the vaginal temperature change, APO decreased the temperature slightly in the vaginal wall while saline administration did not show any temperature change in the vaginal wall. As the outcomes demonstrated, the developed probe can detect hemodynamic and temperature variation in the vaginal wall. The hemodynamic information acquired by the probe can be utilized to establish an objective and accurate standard of female sexual disorders.

Two genetic linkage maps, constructed by DH and RILs populations derived from the same parents, were carried out for the identification and comparison of QTLs controlling yield traits across different years in rice (Oryza sativa L.). A total of 194 SSR and STS markers were used in two maps, of which 114 markers were same. The distribution of Samgang allele was higher in RILs population than it in DH population. Comparing with DH population, RILs population has more lines with higher yield and wider phenotypic transgressive segression for yield traits. Although most of QTLs for the same trait were different in two populations across different years, 8 QTLs (including gwp11.1, spp5.1, spp10.1, spp11.2, ssr1.1, ssr11.1, tgw9.1 and tgw11.1) were detected over 2 yr. It is important to note that ppp10.1, spp10.1 and tgw9.1 were identified in two populations, while spp10.1 and tgw9.1 were simultaneity observed across different years. Epistatic effects were more important than additive effects for PPP, SPP, yield in DH population and TGW, yield in RILs population. Epistatic effects of DH and RILs populations were different on the same genetic background in the present study, which illuminated the QE interaction played an important role on epistatic effect. Identification and comparison of QTLs for yield traits in DH and RILs populations should provide various and more precise information. The QTLs identified in present study would be valuable in marker-assisted selection program for improving rice yield.