Search

Global food insecurity is expected to worsen in the coming decades with the accelerated rate of climate change and the rapidly increasing population. In this vein, it is important to remove inefficiencies at every level of food production. The recent advances in deep learning can help reduce such inefficiencies, yet their application has not yet become mainstream throughout the industry, inducing economic costs at a massive scale. To this point, modern techniques such as CNNs (Convolutional Neural Networks) have been applied to RPQD (Raw Produce Quality Detection) tasks. On the other hand, Transformer's successful debut in the vision among other modalities led us to expect a better performance with these Transformer-based models in RPQD. In this work, we exclusively investigate the recent state-of-the-art Swin (Shifted Windows) Transformer which computes self-attention in both intra- and inter-window fashion. We compare Swin Transformer against CNN models on four RPQD image datasets, each containing different kinds of raw produce: fruits and vegetables, fish, pork, and beef. We observe that Swin Transformer not only achieves better or competitive performance but also is data- and compute-efficient, making it ideal for actual deployment in real-world setting. To the best of our knowledge, this is the first large-scale empirical study on RPQD task, which we hope will gain more attention in future works.

Data scarcity and noise are important issues in industrial applications of machine learning. However, it is often challenging to devise a scalable and generalized approach to address the fundamental distributional and semantic properties of dataset with black box models. For this reason, data-centric approaches are crucial for the automation of machine learning operation pipeline. In order to serve as the basis for this automation, we suggest a domain-agnostic pipeline for refining the quality of data in image classification problems. This pipeline contains data valuation, cleansing, and augmentation. With an appropriate combination of these methods, we could achieve 84.711% test accuracy (ranked #6, Honorable Mention in the Most Innovative) in the Data-Centric AI competition only with the provided dataset., Comment: Data Centric AI Workshop at NeurIPS 2021

In the present study, an amplifying neuron and attenuating neuron, which can be easily implemented into neural networks without any significant additional computational effort, are proposed. The activated output value is squared for the amplifying neuron, while the value becomes its reciprocal for the attenuating one. Theoretically, the order of neural networks increases when the amplifying neuron is placed in the hidden layer. The performance assessments of neural networks were conducted to verify that the amplifying and attenuating neurons enhance the performance of neural networks. From the numerical experiments, it was revealed that the neural networks that contain the amplifying and attenuating neurons yield more accurate results, compared to those without them., 8 pages, the figure 6-(b) and (c) were exchanged

Receptor-mediated endocytosis is essential for targeted gene/drug delivery to a specific cell type. In this study, we developed a capacitance sensor to monitor receptor-mediated endocytosis in real time. The capacitance sensor was able to detect a capacitance peak in different cell lines during the internalization of adenoviruses or antibodies via receptor-mediated endocytosis. In contrast, the capacitance declined without a capacitance peak when nanoparticles were taken up via non-specific pinocytosis. Thus, our capacitance sensor represents a potential capacitance-based means of discrimination between receptor-mediated endocytosis and non-specific pinocytosis. Moreover, we developed a capacitance sensor array to demonstrate capacitance-based high-throughput screening. We showed that the capacitance sensor array could rapidly identify antibodies or ligands with high specificity for target molecules. We propose that the capacitance sensor array will provide a valuable tool for high-throughput screening.

Numerical simulations of the NREL phase VI wind turbine operating in yawed conditions have been performed using two computational methods; one based on the solution of the Reynolds-averaged Navier-Stokes equations (RANS) using unstructured overset meshes and one known as the Vorticity Transport Model (VTM) that is based on the solution of the vorticity transport equation. Simulations with and without the instrumentation enclosure that was present during the NREL experiments were performed. It was found that the enclosure influenced the loading significantly at the inboard part of the blade when the blade passed through the wake that was developed by the enclosure. Both the RANS and VTM codes are able to predict well the unsteady and time-averaged aerodynamic loadings on the wind turbine blades at low wind speeds. At high wind speeds, leading-edge flow separation and strong radial flow are observed on the suction surface of the blades, when the blades are at the retreating side of the rotor. Both the RANS and VTM codes provide less accurate predictions of the blade loads. However, at the advancing side of the rotor, the flow is mostly attached to the surface of the blade, and both the RANS and VTM predictions of the blade loads are in good agreement with the measured data.

Numerical simulations of the NREL phase VI wind turbine operating in yawed conditions have been performed using two computational methods; one based on the solution of the Reynolds-averaged Navier-Stokes equations (RANS) using unstructured overset meshes and one known as the Vorticity Transport Model (VTM) that is based on the solution of the vorticity transport equation. The effect of the hub that was present during the NREL experiments was investigated by modeling the hub in the RANS simulations. It was found that the hub influenced the loading significantly at the inboard part of the blade when the blade passed through the wake that was developed by the hub. Both the RANS and VTM codes are able to predict well the unsteady and time-averaged aerodynamic loadings on the wind turbine blades at low wind speeds. At high wind speeds, leading-edge flow separation and strong radial flow are observed on the suction surface of the blades, when the blades are at the retreating side of the rotor. Both the RANS and VTM codes provide less accurate predictions of the blade loads. However, at the advancing side of the rotor, the flow is mostly attached to the surface of the blade, and both the RANS and VTM predictions of the blade loads are in good agreement with the measured data.

Two computational methods, one based on the solution of the vorticity transport equation, and a second based on the solution of the Reynolds-Averaged Navier-Stokes equations, have been used to simulate the aerodynamic performance of a horizontal axis wind turbine. Comparisons have been made against data obtained during Phase VI of the NREL Unsteady Aerodynamics Experimental and against existing numerical data for a range of wind conditions. The Reynolds-Averaged Navier-Stokes method demonstrates the potential to predict accurately the flow around the blades and the distribution of aerodynamic loads developed on them. The Vorticity Transport Model possesses a considerable advantage in those situtations where the accurate, but computationally efficient, modelling of the structure of the wake and the associated induced velocity is critical, but where the prediction of blade loads can be achieved with sufficient accuracy using a lifting-line model augmented by incorporating a semi-empirical stall delay model. The largest benefits can be extracted when the two methods are used to complement each other in order to understand better the physical mechanisms governing the aerodynamic performance of wind turbines.

The biodegradable inorganic nanovector based on a layered double hydroxide (LDH) holds great promise for gene and drug delivery systems. However, in vivo targeted delivery of genes through LDH still remains a key challenge in the development of RNA interference therapeutics. Here, we describe in vivo and in vitro delivery system for Survivin siRNA (siSurvivin) assembled with passive LDH with a particle size of 100 nm or active LDH conjugated with a cancer overexpressing receptor targeting ligand, folic acid (LDHFA), conferring them an ability to target the tumor by either EPR-based clathrin-mediated or folate receptor-mediated endocytosis. When not only transfected into KB cells but also injected into xenograft mice, LDHFA/siSurvivin induced potent gene silencing at mRNA and protein levels in vitro, and consequently achieved a 3.0-fold higher suppression of tumor volume than LDH/siSurvivin in vivo. This anti-tumor effect was attributed to a selectively 1.2-fold higher accumulation of siSurvivin in tumor tissue compared with other organs. Targeting to the tumor with inorganic nanovector can guide and accelerate an evolution of next-generation theranosis system. Refereed/Peer-reviewed