Search

We propose PyCAN, the first open-source Python implementation of N-dimensional Content-Addressable Network (CAN) with full feature sets to maintain peer-to-peer structure. Existing CAN implementations supports limited functions of Distributed Hash Table (DHT), so they cannot be used in practice. However, PyCAN offers full set of features such as N-dimension coordinates, node removal followed by node taking over, and a verification method to keep peer-to-peer structure. By extensive experiments, we confirm that PyCAN supports scalable overlay structures, so it is practically usable in peer-to-peer systems.

Although plant disease recognition has witnessed a significant improvement with deep learning in recent years, a common observation is that current deep learning methods with decent performance tend to suffer in real-world applications. We argue that this illusion essentially comes from the fact that current plant disease recognition datasets cater to deep learning methods and are far from real scenarios. Mitigating this illusion fundamentally requires an interdisciplinary perspective from both plant disease and deep learning, and a core question arises. What are the characteristics of a desired dataset? This paper aims to provide a perspective on this question. First, we present a taxonomy to describe potential plant disease datasets, which provides a bridge between the two research fields. We then give several directions for making future datasets, such as creating challenge-oriented datasets. We believe that our paper will contribute to creating datasets that can help achieve the ultimate objective of deploying deep learning in real-world plant disease recognition applications. To facilitate the community, our project is publicly available at https://github.com/xml94/PPDRD with the information of relevant public datasets.

Abstract A new perspective suggests that a dynamic bidirectional communication system, often referred to as the microbiome-gut-brain axis, exists among the gut, its microbiome, and the central nervous system (CNS). This system may influence brain health and various brain-related diseases, especially in the realms of neurodevelopmental and neurodegenerative conditions. However, the exact mechanism is not yet understood. Metabolites or extracellular vesicles derived from microbes in the gut have the capacity to traverse the intestinal epithelial barrier or blood–brain barrier, gaining access to the systemic circulation. This phenomenon can initiate the physiological responses that directly or indirectly impact the CNS and its function. However, reliable and controllable tools are required to demonstrate the causal effects of gut microbial-derived substances on neurogenesis and neurodegenerative diseases. The integration of microfluidics enhances scientific research by providing advanced in vitro engineering models. In this study, we investigated the impact of microbe-derived metabolites and exosomes on neurodevelopment and neurodegenerative disorders using human induced pluripotent stem cells (iPSCs)-derived neurons in a gut-brain axis chip. While strain-specific, our findings indicate that both microbial-derived metabolites and exosomes exert the significant effects on neural growth, maturation, and synaptic plasticity. Therefore, our results suggest that metabolites and exosomes derived from microbes hold promise as potential candidates and strategies for addressing neurodevelopmental and neurodegenerative disorders.

Arthritis is mainly a geriatric disease that causes joint pain and lowers the quality of life. This clinical trial was performed to evaluate the efficacy of Lilium lancifolium Thunb. (HY-LL) in alleviating joint pain. Six candidate anti-inflammatory components including regaloside A were identified in HY-LL using HPLC analysis. All participants were assigned to the HY-LL or the placebo group and took tablets twice a day for 12 weeks. As a result, pain VAS and K-WOMAC total scores significantly decreased after 12 weeks compared to the baseline in the HY-LL group, with a statistically significant difference between the two groups (p = 0.043, 0.043). The K-WOMAC sub-scores for pain and function showed a statistically significant improvement in the HY-LL group compared to the placebo group (p = 0.023, 0.047). Furthermore, the participants’ overall quality of life improved after 12 weeks of HY-LL consumption (p = 0.024). However, no significant differences were observed in the blood biomarkers. Therefore, this study demonstrated the positive effect of 12 weeks of HY-LL consumption on joint pain and quality of life.

Abstract Transition metal dichalcogenides (TMDs) have emerged as a promising alternative to noble metals in the field of electrocatalysts for the hydrogen evolution reaction. However, previous attempts using machine learning to predict TMD properties, such as catalytic activity, have been shown to have limitations in their dependence on large amounts of training data and massive computations. Herein, we propose a genetic descriptor search that efficiently identifies a set of descriptors through a genetic algorithm, without requiring intensive calculations. We conducted both quantitative and qualitative experiments on a total of 70 TMDs to predict hydrogen adsorption free energy ( $$\Delta G_H$$ Δ G H ) with the generated descriptors. The results demonstrate that the proposed method significantly outperformed the feature extraction methods that are currently widely used in machine learning applications.

In distributed deep learning, the improper use of the collective communication library can lead to a decline in deep learning performance due to increased communication time. Representative collective communication libraries such as MPI, GLOO, and NCCL exhibit varying performance based on server environment and communication architecture. In this study, we investigate three key aspects to evaluate the performance of the collective communication libraries in a distributed deep learning setting in an intra-node environment. First, we conduct a comparison and analysis of collective communication library performance within common distributed deep learning architectures, such as parameter servers and ring all-reduce methods. Second, we evaluate the performance of these libraries in different environments, including various container platforms and bare metal setups, considering the scalability and flexibility advantages offered by cloud virtualization. Last, to ensure practicality, we assess the libraries’ performance in a Linux shell and within the PyTorch framework. In the cross-docker virtualization environment, NCCL shows up to 213% higher latency compared to single docker, while GLOO exhibits 36% lower latency in single docker than in cross docker, and NCCL achieves up to 345% lower execution time in all-reduce operations compared to other libraries (MPI and GLOO). These findings will inform the selection of an appropriate collective communication library for designing effective distributed deep learning environments.

This article presents a high-performance, low-power analog convolutional neural network (CNN) circuit integrated into a CMOS image sensor (CIS) for face detection applications. The main block of the proposed in-column analog CNN circuits is an analog multiplication-and-accumulation (MAC) circuit consisting of an operational transconductance amplifier-based switched capacitor circuit enabling the programmable weight function. With the proposed MAC, a 3-layer analog CNN processor is implemented into the column-parallel readout circuit in conventional CIS. Furthermore, for low-power CNN operations, we use a low-resolution analog-to-digital converter with the proposed nonlinear quantization method resulting in an increase in the accuracy of face detection from 92.8% to 98.75% at 120 frame rates with 2.8 V/1.5 V supply voltage. A prototype sensor with $160\times120$ effective image resolution was fabricated using a 110 nm CMOS image sensor process. The measurement results showed that the maximum power consumption was 0.57 mW and 4.02 mW at 1 and 120 frame rates, respectively.

In this study, the temperature-dependent electrical characteristics of p-channel mode feedback field-effect transistors (FBFETs) were examined at temperatures ranging from 250 to 425 K. Their steep subthreshold swings of less than 1 mV/dec were maintained even at temperatures up to 400 K. As the temperature increased to 400 K, the latch-up voltage shifted from −0.951 to −0.613 V, which was caused by a reduction in the potential barriers in the channels of the FBFETs. High $I_{\mathrm {on}}/I_{\mathrm {off}}$ ratios above 108 were maintained in the temperature range of 250 to 400 K. However, at temperatures over 400 K, the FBFETs were turned on regardless of the gate voltages owing to the generation of a thermally induced positive feedback loop.

Object detection models have become the current tool of choice for plant disease detection in precision agriculture. Most existing research improved the performance by ameliorating networks and optimizing the loss function. However, because of the vast influence of data annotation quality and the cost of annotation, the data-centric part of a project also needs more investigation. We should further consider the relationship between data annotation strategies, annotation quality, and the model’s performance. In this paper, a systematic strategy with four annotation strategies for plant disease detection is proposed: local, semi-global, global, and symptom-adaptive annotation. Labels with different annotation strategies will result in distinct models’ performance, and their contrasts are remarkable. An interpretability study of the annotation strategy is conducted by using class activation maps. In addition, we define five types of inconsistencies in the annotation process and investigate the severity of the impact of inconsistent labels on model’s performance. Finally, we discuss the problem of label inconsistency during data augmentation. Overall, this data-centric quantitative analysis helps us to understand the significance of annotation strategies, which provides practitioners a way to obtain higher performance and reduce annotation costs on plant disease detection. Our work encourages researchers to pay more attention to annotation consistency and the essential issues of annotation strategy. The code will be released at: https://github.com/JiuqingDong/PlantDiseaseDetection_Yolov5 .

Audio auto-tagging is the process of assigning labels to audio clips for better categorization and management of audio file databases. With the advent of advanced artificial intelligence technologies, there has been increasing interest in directly using raw audio data as input for deep learning models in order to perform tagging and eliminate the need for preprocessing. Unfortunately, most current studies of audio auto-tagging cannot effectively reflect the semantic relationships between tags—for instance, the connection between “classical music” and “cello”. In this paper, we propose a novel method that can enhance audio auto-tagging performance via joint embedding. Our model has been carefully designed and architected to recognize the semantic information within the tag domains. In our experiments using the MagnaTagATune (MTAT) dataset, which has high inter-tag correlations, and the Speech Commands dataset, which has no inter-tag correlations, we showed that our approach improves the performance of existing models when there are strong inter-tag correlations.

Abstract Efforts to improve CRISPR-Cas9 genome editing systems for lower off-target effects are mostly at the cost of its robust on-target efficiency. To enhance both accuracy and efficiency, we created chimeric SpyCas9 proteins fused with the 5′-to-3′ exonuclease Recombination J (RecJ) or with GFP and demonstrated that transfection of the pre-assembled ribonucleoprotein of the two chimeric proteins into human or plant cells resulted in greater targeted mutagenesis efficiency up to 600% without noticeable increase in off-target effects. Improved activity of the two fusion proteins should enable editing of the previously hard-to-edit genes and thus readily obtaining the cells with designer traits.

Mobile devices such as smartphones and tablets have become widespread, and studies are being conducted to improve the convenience of smartphone users. The I/O performance is considered an important factor affecting the quality of the smartphone user experience. Therefore, methods for retaining applications with a high launch frequency in the main memory, improving the I/O stack of the smartphone operating system, and a new file system for improving the I/O performance have been actively studied. However, there is no information sharing system between the I/O stacks in a smartphone operating system. Existing studies suffer from limitations in improving the file I/O performance because the upper- and lower-layer information of the I/O stack was not considered simultaneously. In this paper, we propose a context-aware file I/O management system (CAFIO) for analyzing context information collected from various layers of the I/O stack and sharing them between the layers. CAFIO collects smartphone application usage patterns and combines them with the lower-layer I/O information to improve the launch speed and I/O latency of an application. CAFIO exhibits average improvements of 26% in application launch time and 45% in cache hit ratio, as well as a read speed 63% faster than that of existing file I/O management systems.

Instance-level image translation aims to only translate instance of interest and can be operated more finely and flexibly than object-level and holistic-level image translation. However, current algorithms are not suitable to do it since they employ a holistic or object level’s discriminator that tends to change the whole image or all instances. To address the issue, we propose a simple yet effective local discriminator, in which the input image is split into two parts, region of interest (ROI) and background. Instance mask is employed to align the ROI and the background is design to be random in a prior distribution to mitigate a divergence between the ROI and the background. In this way, we obtain translated instance with decent margins without artifacts as current algorithms get. Moreover we propose a new architecture to simultaneously realize versatile instance-level image translation. Experimental results prove that our proposed algorithm outperforms the state-of-the-art in position accuracy and background retainment by a clear margin.

In this study, we demonstrated the enhancement of the output power of a hybrid energy device (HED) using a cooling patch that does not consume any external electric power. The HED consisted of a photovoltaic cell (PVC) and a thermoelectric generator (TEG); the cooling patch was attached to the TEG. When the PVC was exposed to solar irradiance, the cooling patch lowered the temperature of the PVC and increased the thermal gradient across the TEG, thereby increasing the output power. For an HED with a cooling patch at an irradiance of 1000 W/m2, the output power increased to 24.2 mW, as compared to the output power of 19.9 mW for an HED without any cooling patch.

One of the key components of the intelligent transport system is to provide a safe driving environment. The ability to quickly and accurately recognize the surrounding environments and neighboring vehicles is essential for the development of safety-related services to provide a safe driving environment. Currently, research in this field is being conducted based on the global positioning system (GPS), light detection and ranging, camera, and ranging sensors. However, the currently used sensors cannot recognize a wide range of vehicles because of their limited range of recognition. Moreover, GPS-based studies are highly affected by the surrounding environment because of the nature of GPS and have relatively high error rates and low rate of location information updates. The use of GPS-based location information for safety-related services can result in negative consequences. In this paper, we propose a new positioning system, called cooperative neighboring vehicle positioning system (CNVPS). The CNVPS rapidly identifies the locations of neighboring vehicles based on their information obtained through various sensors and shares this information with a wide range of neighboring vehicles over vehicle-to-vehicle communications. The CNVPS also compensates the position of a neighboring vehicle by applying the maximum likelihood estimation to the duplicated position observation of the other neighboring vehicles. The simulation results show that the CNVPS achieves approximately 370% improvement in location errors over GPS, assuming that the root-mean-squared error of GPS is 15 m. In addition, the proposed system has a location refresh cycle ten times faster than the existing GPS-based system.

Smartphones that are equipped with high-clock frequency and multi-core processors are being commercially released to provide various services. As the number of cores and the clock speed of a mobile processor increases, its power consumption also increases, and several software approaches to reducing power consumption have been studied. Existing techniques estimate processor usage by measuring the processor usage at a previous time. However, these techniques often waste energy because they assign frequencies above the usage required to prevent degraded user responsiveness. Therefore, this paper proposes a machine learning method to predict the usage that the processor currently requires to prevent performance degradation while reducing power consumption. The proposed method is implemented through a processor power management system based on Long Short-Term Memory (LSTM). This system learns processor usage patterns in a variety of situations and predicts the processor usage required for the current situation. The number of computations required by the LSTM-based technique is analyzed according to the number of neurons and layers, and the computational load is then compared to an existing technique. Furthermore, a benchmarking tool that reflects the characteristics of mobile applications is used to test the performance of the proposed system, which is shown to reduce the power consumption of mobile processors by a maximum of 19% compared to the existing Android processor power management system.

This study was to determine the effect of oleogel on lipid metabolism in rats fed high-fat diet. Male SD rats (n = 32; 4-wk-old) and obese Zucker rats (n = 6) fed either beef tallow containing high fat diet (HFD) or beeswax oleogel containing canola oil (BOG) for 2 months. Substitution of 50% BT by BOG significantly reduced body weight. Increased serum triglyceride by HFD was reduced (P

Background: Explosive puffing can induce changes in the chemical, nutritional, and sensory quality of red ginseng. The antioxidant properties of ethanolic extracts of red ginseng and puffed red ginseng were determined in bulk oil and oil-in-water (O/W) emulsions. Methods: Bulk oils were heated at 60°C and 100°C and O/W emulsions were treated under riboflavin photosensitization. In vitro antioxidant assays, including 2,2-diphenyl-1-picrylhudrazyl, 2,2′-azinobis-3-ethyl-benzothiazoline-6-sulfonic acid, ferric reducing antioxidant power, total phenolic content, and total flavonoid content, were also performed. Results: The total ginsenoside contents of ethanolic extract from red ginseng and puffed red ginseng were 42.33 mg/g and 49.22 mg/g, respectively. All results from above in vitro antioxidant assays revealed that extracts of puffed red ginseng had significantly higher antioxidant capacities than those of red ginseng (p

As a way to increase the actual main memory capacity of Android smartphones, most of them make use of zRAM swapping, but it has limitation in increasing its capacity since it utilizes main memory. Unfortunately, they cannot use secondary storage as a swap space due to the long response time and wear-out problem. In this paper, we propose a hybrid swapping scheme based on per-process reclaim that supports both secondary-storage swapping and zRAM swapping. It attempts to swap out all the pages in the working set of a process to a zRAM swap space rather than killing the process selected by a low-memory killer, and to swap out the least recently used pages into a secondary storage swap space. The main reason being is that frequently swapin/out pages use the zRAM swap space while less frequently swap-in/out pages use the secondary storage swap space, in order to reduce the page operation cost. Our scheme resolves both the response time and wear-out problems of secondary-storage swapping and zSWAP, and overcomes the size limitation of the zRAM swap space. According to performance measurements, it also increased the extension ratio of main memory by 15 17% and 6 17% and reduced the page operation cost by 9 22% and 18 28%, respectively, compared with zRAM swapping and zSWAP.

To achieve high accuracy when performing deep learning, it is necessary to use a large-scale training model. However, due to the limitations of GPU memory, it is difficult to train large-scale training models within a single GPU. NVIDIA introduced a technology called CUDA Unified Memory with CUDA 6 to overcome the limitations of GPU memory by virtually combining GPU memory and CPU memory. In addition, in CUDA 8, memory advise options are introduced to efficiently utilize CUDA Unified Memory. In this work, we propose a newly optimized scheme based on CUDA Unified Memory to efficiently use GPU memory by applying different memory advise to each data type according to access patterns in deep learning training. We apply CUDA Unified Memory technology to PyTorch to see the performance of large-scale learning models through the expanded GPU memory. We conduct comprehensive experiments on how to efficiently utilize Unified Memory by applying memory advises when performing deep learning. As a result, when the data used for deep learning are divided into three types and a memory advise is applied to the data according to the access pattern, the deep learning execution time is reduced by 9.4% compared to the default Unified Memory.

In this paper, we present several optimization strategies that can improve the overall performance of the distributed in-memory computing system, “Apache Spark”. Despite its distributed memory management capability for iterative jobs and intermediate data, Spark has a significant performance degradation problem when the available amount of main memory (DRAM, typically used for data caching) is limited. To address this problem, we leverage an SSD (solid-state drive) to supplement the lack of main memory bandwidth. Specifically, we present an effective optimization methodology for Apache Spark by collectively investigating the effects of changing the capacity fraction ratios of the shuffle and storage spaces in the “Spark JVM Heap Configuration” and applying different “RDD Caching Policies” (e.g., SSD-backed memory caching). Our extensive experimental results show that by utilizing the proposed optimization techniques, we can improve the overall performance by up to 42%.

Recently, cyber-physical system is widely used for smart system control in various fields. Various functions of the cyber-physical system must overcome the limited hardware resources constraint of an embedded system. In addition, the data required from the industrial cyber-physical system are critical; therefore, a highly secure encryption technique is required. However, security and computational throughput are incompatible with each other in the cryptographic technique; therefore, the industrial cyber-physical system needs to adopt a highly efficient and secure encryption technique considering the limited available resources. This study applies the m-folding method to the highly secure elliptic curve algorithm to improve efficiency and proposes the cryptosystem optimized for the resource-constrained industrial cyber-physical system. The proposed m-folding method–based elliptic curve encryption showed 50% faster encryption than the existing methods.

In a distributed deep learning system, a parameter server and workers must communicate to exchange gradients and parameters, and the communication cost increases as the number of workers increases. This paper presents a communication data optimization scheme to mitigate the decrease in throughput due to communication performance bottlenecks in distributed deep learning. To optimize communication, we propose two methods. The first is a layer dropping scheme to reduce communication data. The layer dropping scheme we propose compares the representative values of each hidden layer with a threshold value. Furthermore, to guarantee the training accuracy, we store the gradients that are not transmitted to the parameter server in the worker’s local cache. When the value of gradients stored in the worker’s local cache is greater than the threshold, the gradients stored in the worker’s local cache are transmitted to the parameter server. The second is an efficient threshold selection method. Our threshold selection method computes the threshold by replacing the gradients with the L1 norm of each hidden layer. Our data optimization scheme reduces the communication time by about 81% and the total training time by about 70% in a 56 Gbit network environment.

To accommodate lots of training data and complex training models, “distributed” deep learning training has become employed more and more frequently. However, communication bottlenecks between distributed systems lead to poor performance of distributed deep learning training. In this study, we proposed a new collective communication method in a Python environment by utilizing Multi-Channel Dynamic Random Access Memory (MCDRAM) in Intel Xeon Phi Knights Landing processors. Major deep learning software platforms, such as TensorFlow and PyTorch, offer Python as a main development language, so we developed an efficient communication library by adapting Memkind library, which is a C-based library to utilize high-performance memory MCDRAM. For performance evaluation, we tested the popular collective communication methods in distributed deep learning, such as Broadcast, Gather, and AllReduce. We conducted experiments to analyze the effect of high-performance memory and processor location on communication performance. In addition, we analyze performance in a Docker environment for further relevance given the recent major trend of Cloud computing. By extensive experiments in our testbed, we confirmed that the communication in our proposed method showed performance improvement by up to 487%.

Infrared detectors on satellites and spacecraft require cooling to increase their measurement sensitivity. To efficiently cool infrared detectors in a zero gravity environment and in limited spaces, a cryogenic loop heat pipe (CLHP) can be used to transfer heat over a certain distance by the capillary forces generated from porous wicks without a mechanical power source. The CLHP presented in this study transfers the heat load to a condenser 0.5 m away from an evaporator at temperatures below −150 °C. The CLHP with two evaporators includes a subloop for initial start-up, and uses a pressure reduction reservoir (PRR) for the supercritical start-up from room to cryogenic temperature. Nitrogen is used as the working fluid to verify the thermal behavior of the CLHP, and the heat-transfer capacity according to the nitrogen charging pressure of the PRR is investigated. To simulate a cryogenic environment, the CLHP is installed inside a space environment simulator, including a single-stage GM (Gifford McMahon) cryocooler to cool the condenser. The CLHP is horizontally installed to simulate zero gravity. The heat-transfer characteristics are experimentally evaluated through the loop circulation of the CLHP.

World epidemic obesity is a major contributing factor to metabolic diseases including insulin resistance and cardiovascular diseases. In this study, aqueous and ethanolic extracts of hulled barley with roasting temperatures of up to 250 °C were investigated for their anti-adipogenic effects in vitro and in vivo. An aqueous extract of hulled barley roasted at 210 °C (AHB210) effectively inhibited adipocyte differentiation. Intraperitoneal injections of 15 or 50 mg/kg AHB210 dose dependently prevented body weight gain, fat mass increase, and dysregulated lipid profiles in high fat diet-induced obese male mice. In addition, oral administration of AHB210 to ovariectomized rats also prevented body weight gain. A high performance liquid chromatographic analysis identified coumaric acid and ferulic acid as primary anti-obesity mediators. The presence of beta glucan in AHB210 was less likely to be responsible for the lipid accumulating actions. Taken together, AHB210 may be useful to prevent obesity and its related metabolic diseases.

Bee venom contains a number of pharmacologically active components, including enzymes and polypeptides such as phospholipase A2 (PLA2) and melittin, which have been shown to exhibit therapeutic benefits, mainly via attenuation of inflammation, neurotoxicity, and nociception. The individual components of bee venom may manifest distinct biological actions and therapeutic potential. In this study, the potential mechanisms of action of PLA2 and melittin, among different compounds purified from honey bee venom, were evaluated against Parkinson’s disease (PD). Notably, bee venom PLA2 (bvPLA2), but not melittin, exhibited neuroprotective activity against PD in a 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) mouse model of PD. MPTP-induced behavioral deficits were also abolished after bvPLA2 treatment, depending on the PLA2 content. Further, bvPLA2 administration activated regulatory T cells (Tregs) while inhibiting inflammatory T helper (Th) 1 and Th17 cells in the MPTP mouse model of PD. These results indicate that bvPLA2, but not melittin, protected against MPTP and alleviated inflammation in PD. Thus, bvPLA2 is a promising and effective therapeutic agent in Parkinson’s disease.

The reliability and availability of FOODLABfat system for determining acid value (AV) and peroxide value (POV) were assessed during the hydrolytic rancidification and lipid oxidation of edible oils. This reagent kit based on spectrophotometry was compared to the official methods (ISO 660 and 3960 protocols) based on manual titration employing the standard mixture for the simulated oxidation models and edible oils during the thermally induced oxidation at 180°C. The linear regression line of standard mixture and the significant difference of thermally oxidized time course study determined between them showed high correlations (R2=0.998 and p