Search

• Empirical observations on convolutional neural networks are presented. • The size of the receptive field did not guarantee improved accuracy. • We observed that dead pixels exist in modern convolutional neural networks. • The effect of dead pixels is discussed with a solution to resolve them. Deep neural networks have been used in various fields, but their internal behavior in how they understand images is not well known. In this study, we discuss two counterintuitive properties of convolutional neural networks (CNNs). First, we evaluated the size of the receptive field of CNNs with their classification accuracy. Previous studies have attempted to increase the size of the receptive field for performance gain. However, we observed that some CNNs with a smaller receptive field can achieve higher classification accuracy. In this regard, we claim that a larger receptive field does not guarantee improved classification accuracy. Second, using the effective receptive field, we examined the contribution of each pixel to the output of CNN. Intuitively, each pixel is expected to equally contribute to the final output, but we found that there exist pixels in a partially dead state with little contribution to the output. We reveal that the reason for dead pixels lies in even stride operations with odd-sized kernels in CNN and propose a kernel padding method to remove the dead pixels. We demonstrated the vulnerability of CNNs with dead pixels when we detect a noise or small box that is on dead pixels. Our findings on dead pixels should be understood and considered in practical applications of CNN. [ABSTRACT FROM AUTHOR]

This multicenter phase I/IIa study aimed to determine the recommended phase II dose (RP2D) and evaluate the safety and preliminary efficacy of liposomal irinotecan (nal-IRI), oxaliplatin, and S-1 (NASOX) as first-line treatment for advanced pancreatic adenocarcinoma. Patients with locally advanced or metastatic pancreatic adenocarcinoma without prior systemic treatment for advanced disease, aged ≥ 19 years, with measurable disease, and Eastern Cooperative Oncology Group performance status of 0–1 were eligible. The primary endpoints were to determine the dose-limiting toxicity (DLT) in the phase I cohort and overall response rate (ORR) in the phase IIa cohort. The intention-to-treat (ITT) analysis included patients who received the RP2D. In phase I, seven patients were screened, and six were assessed for DLT. None experienced DLT during the first cycle. The RP2D was determined as nal-IRI 50 mg/m2 and oxaliplatin 60 mg/m2 on day 1, S-1 40 mg/m2 twice daily on days 1–7 every 14 days. For the ITT (N = 41; 7, and 34 from phases I and IIa, respectively), the most common grade 3–4 treatment-emergent adverse events were neutropenia (31.7 %), enterocolitis (9.8 %), anorexia (7.3 %), and diarrhea (2.4 %). The ORR was 58.5 % (1 complete, and 23 partial responses). Two underwent conversion surgery; both achieved R0 resection. With median follow-up of 17.5 months, median progression-free survival was 6.5 months (95 % confidence interval [CI], 5.0–8.1) and median overall survival was 11.4 months (95 % CI, 9.8–15.5). NASOX exhibited a manageable safety profile and encouraging efficacy outcomes consistent with NALIRIFOX, showing potential to replace infusional 5-fluorouracil with oral S-1 in the triplet regimen. • In this Ph I/IIa trial, patients with advanced PDAC received 1 L NASOX every 2 weeks. • RP2D: nal-IRI 50 mg/m2, oxaliplatin 60 mg/m2 on D1, S-1 40 mg/m2 BID on D1–7. • ORR was 58.5 %, median PFS was 6.5 months, and median OS was 11.4 months. • Safety profiles were manageable. • NASOX may improve patient convenience while showing efficacy comparable to NALIRIFOX. [ABSTRACT FROM AUTHOR]

Recently, various normalization layers have been proposed to stabilize the training of deep neural networks. Among them, group normalization is a generalization of layer normalization and instance normalization by allowing a degree of freedom in the number of groups it uses. However, to determine the optimal number of groups, trial-and-error-based hyperparameter tuning is required, and such experiments are time-consuming. In this study, we discuss a reasonable method for setting the number of groups. First, we find that the number of groups influences the gradient behavior of the group normalization layer. Based on this observation, we derive the ideal number of groups, which calibrates the gradient scale to facilitate gradient descent optimization. This paper is the first to propose an optimal number of groups that is theoretically grounded, architecture-aware, and can provide a proper value in a layer-wise manner for all layers. The proposed method exhibited improved performance over existing methods in numerous neural network architectures, tasks, and datasets. • We propose a method to determine the number of groups of group normalization. • A theoretical analysis of group normalization with activation function is provided. • The proposed method is validated against various practical problems. [ABSTRACT FROM AUTHOR]

• We provide empirical tests on various image corruption using vision transformers. • Vision transformers showed performance degradation on contrast-enhanced images. • We proposed PreLayerNorm for the consistent behavior of positional embedding. • We observed that PreLayerNorm improved performance on contrast-enhanced images. • We provide theoretical analyses on the inconsistent behavior of vision transformers. [Display omitted] Vision Transformers (ViTs) have recently demonstrated state-of-the-art performance in various vision tasks, replacing convolutional neural networks (CNNs). However, because ViT has a different architectural design than CNN, it may behave differently. To investigate whether ViT has a different performance or robustness, we tested ViT and CNN under various imaging conditions in practical vision tasks. We confirmed that for most image transformations, ViT's robustness was comparable or even better than that of CNN. However, for contrast enhancement, ViT performed particularly poorly. We show that this is because positional embedding in ViT's patch embedding can work improperly when the color scale changes. We demonstrate that the use of PreLayerNorm, a modified patch embedding structure, ensures the consistent behavior of ViT. Results demonstrate that ViT with PreLayerNorm exhibited improved robustness in the contrast-varying environments. [ABSTRACT FROM AUTHOR]

Predicting electrical power consumption is essential in many industries. This paper presents a method to predict total electrical power consumption in a certain period of Hot Rolling Mill (HRM) in steel mills. We adequately pre-processed operation data, including work schedule, work result, and power measurement. In addition, based on domain knowledge, we derived several useful variables relevant to power consumption. To predict power consumption, we developed a Multi-Level Stacked Regression model that combines the strengths of multiple prediction models. Real-world applications have dataset shift characteristics in which the covariance between features or distribution of the features changes over time due to frequent changes in the working environment. To verify the generalization accuracy of our model, we trained and tested our model for various training/test splits and compared it with several comparative models. Our model showed the best accuracy in all training/test splits. We verified whether the predictive model reflects the real-world environment well by analyzing the actual physical meaning of the essential features used in the predictive model. • A series of pre-processing methods are applied to real data from Hot Rolling Mill. • A Multi-Level Stacked Regression algorithm is proposed. • Performance on various training/test splits that contains dataset shifts is verified. [ABSTRACT FROM AUTHOR]

Go, a heterotrimeric G-protein, is enriched in brain and neuronal growth cones. Although several reports suggest that Go may be involved in modulation of neuronal differentiation, the precise role of Go is not clear. To investigate the function of Go in neuronal differentiation, we determined the effect of Goα, the α subunit of Go, on the expression of Cav2.2, the pore-forming unit of N-type calcium channels, at the transcription level. Treatment with cyclic AMP (cAMP), which triggers neurite outgrowth in neuroblastoma F11 cells, increased the mRNA level and the promoter activity of the Cav2.2 gene. Overexpression of Goα inhibited neurite extension in F11 cells and simultaneously repressed the stimulatory effect of cAMP on the Cav2.2 gene expression to the basal level. Targeted mutation of the Goα gene also increased the level of Cav2.2 in the brain. These results suggest that Go may regulate neuronal differentiation through modulation of gene expression of target genes such as N-type calcium channels. [Copyright &y& Elsevier]

Layer-by-layer etching of Si(1 0 0) has been investigated by alternating chlorine adsorption and Ar+ ion irradiation. As the Ar+ ion source, a helicon plasma has been used to invoke the surface reaction of Si with the adsorbed chlorine. At a chlorine partial pressure of 10 mPa, the Si surface was found to be fully saturated with adsorbed chlorine in 20 s. In order to achieve the layer-by-layer etching of Si, the Ar+ ion acceleration energy should be above 20 eV at the time of chlorine adsorption. At energies above 40 eV, however, sputtering of Si was found to occur. The self-limited etch rate was measured to be half monolayer per cycle, which was 0.68 A˚ per cycle. The roughness of Si surface was found to increase during the etching by about 22% after 900 cycles. [Copyright &y& Elsevier]

A room-temperature (∼26 °C) NO 2 gas sensor was fabricated using tellurium (Te) nanostructures. Te forms a helical chain with strong covalent bonds in the c-axis and weak van der Waals bonds between the chains. Therefore, the energy required for gas adsorption and desorption is low, thus, giving it the advantage of operating at room temperature. Since the shape and size of the Te helical chains can be freely controlled by adjusting the growth parameters, the electrical reactivities of Te to a specific gas can be controlled. In this study, the reactivity of Te to NO 2 gas was controlled by adjusting the number of Te nanotube walls and lengths. It was also verified that the gas detection performance was maintained even with increasing humidity. [ABSTRACT FROM AUTHOR]

Abstract: GaInP/AlGaInP multiple quantum-well (MQW) laser diodes with diffusion barriers for Zn, grown by metal organic chemical vapor deposition (MOCVD), were fabricated and characterized. To determine the optimal conditions for the diffusion barriers, we investigated Zn diffusion in the GaInP/(Al0.5Ga0.5)0.5In0.5P MQW layers against three undoped (Al0.7Ga0.3)0.5In0.5P diffusion barriers with thicknesses of 0, 90, and 130nm, respectively, by secondary ion mass spectroscopy (SIMS). Compared to the reference AlGaInP laser diode without a diffusion barrier, the AlGaInP ridge laser with a 130nm thick diffusion barrier on the top of the MQW layer had a threshold current that was greatly reduced from 110 to 75mA and the characteristic temperature T o that was increased from 113 to 144K in the temperature range from 25 to 50°C. This report presents quantitative information on the effect of Zn diffusion on laser performance. [Copyright &y& Elsevier]

• A weakly supervised power line detection algorithm is proposed. • A line specialized refinement algorithm is applied for performance improvement. • A repeated learning framework with noise label develops the model output. Detection of power lines in aerial images is an important problem to prevent accidents of unmanned aerial vehicles operating at low altitudes in the electrical industry. Recently, pixel-level power line detection using deep learning has been studied but production of the pixel-level annotations for massive dataset is difficult. In this study, we propose a power line detection algorithm using weakly supervised learning method to reduce the labeling cost for dataset generation. The algorithm is divided into two stages. First, an approximately localized mask was generated based on a convolutional neural network which was trained with only patch-level labels. Second, recursive training of segmentation network with refined broken line segments was executed. A refinement algorithm, line segment connecting (LSC) is a power-line-specialized refinement module that connects broken lines by approximating the segments as partially straight. In proposed algorithm, predicted image at each recursive step was updated as a label of the next training and the label was developed by itself with LSC. The comprehensive experimental results of our algorithm showed state-of-art F 1 -score of 94.3% in weakly supervised learning approaches on public dataset. This result suggests that the proposed algorithm is useful for low labeling cost with high performance in line detection application. [ABSTRACT FROM AUTHOR]

By using a MoI 2 precursor, a one-dimensional linear chain material, Mo 6 S 3 I 6 , was synthesized on a centimeter scale. As MoI 2 can be used to form Mo 6 S 3 I 6 through a gas-phase reaction, a larger amount of single crystals can be synthesized; moreover, these crystals are significantly longer than those synthesized by using Mo, S, and I powder. The synthesized Mo 6 S 3 I 6 exhibited the characteristics of paramagnetic semiconductors with excellent thermal stability and a low bandgap. Furthermore, Mo 6 S 3 I 6 can be separated into long nanometer-scale chains by a solution dispersion process. • First synthesis of cm scale synthesize of vdW 1D material Mo 6 S 3 I 6. • Superior material properties based on high crystallinity. • Atomic level (nm) dispersion for ideal 1D materials. [ABSTRACT FROM AUTHOR]

A new of 2D transition metal thiophosphates, Nb 4 P 2 S 21 , was synthesized through a vapor transport reaction. High purity and centimeter-scale large Nb 4 P 2 S 21 crystals were obtained by controlling the growth temperature and stoichiometric ratio between niobium, phosphorus, and sulfur (using excess sulfur condition, Nb: P: S = 4: 2: 23). It was confirmed that large Nb 4 P 2 S 21 crystals could be separated to a nanometer scale through mechanical exfoliation, and the exfoliated Nb 4 P 2 S 21 layer exhibits semiconducting characteristics by measuring the change in electrical conductivity with temperature (increased conductivity with temperature in the range of 93–333 K). Furthermore, the device characteristics for detecting visible ray laser (635 nm) and NO 2 gas molecules were verified. Based on these characteristics, the synthesized 2D material is expected to be a potential candidate for additional 2D materials and in multi-functional sensor devices, including optoelectronics and gas sensing applications. • Synthesis of 2D transition metal thiophosphates Nb 4 P 2 S 21. • Bulk to few nanometer scale separation of Nb 4 P 2 S 21 using mechanical exfoliation. • Multi-functional sensing application in photodetector and gas sensor. [ABSTRACT FROM AUTHOR]

A new type of 1D inorganic material, MoI 3 , which can be separated into individual 1D chains, was synthesized through a vapor transport reaction between molybdenum and iodine. High-purity and centimeter-scale large MoI 3 crystals were obtained by controlling the growth temperature and the stoichiometric ratio between molybdenum and iodine. Their semiconductor characteristics were confirmed by measuring the electrical conductivity as a function of temperature (conductivity increased with increased temperature across the range −70 to 30 °C). Furthermore, it was confirmed that MoI 3 with a 1.27 eV band gap can be separated to a few nanometer scale through mechanical and chemical exfoliation methods and can be used as a photodetector in the near IR (less than 1000 nm)-to-visible light region. This novel 1D material is expected to play an important role in the development of additional low-dimensional materials and in advanced nano-optoelectronic devices. • Synthesis of 1D inorganic material MoI 3 via chemical vapor transport (CVT) method. • Experimental estimation of band gap (∼ 1.27 eV) and application to photodetector. • Bulk to few nanometer scale separations of MoI 3 chains using mechanical and chemical exfoliation. [ABSTRACT FROM AUTHOR]

The two-dimensional (2D) inorganic material Nb 3 I 8 , which can be separated into atomic planes, was synthesized through a vapor transport reaction between niobium and iodine. High-purity and cm scale large Nb 3 I 8 crystals were obtained by controlling the growth temperature (600–700 °C) and the stoichiometric ratio between niobium and iodine (using excess I 2 condition, niobium: iodine = 3:9). The semiconductor characteristics were measured by the change of electrical conductivity with temperature (increased conductivity with temperature in the range of 100 K–300 K), and have a band gap of ∼1 eV (∼1200 nm) through Infrared (IR)-visible light absorption analysis. Based on these characteristics, the device characteristics of detecting IR near 1200 nm region were verified. Besides few-layers of Nb 3 I 8 was acquired by the mechanical exfoliation method, and experimentally observed by atomic force microscopy (AFM). This 2D material is expected to have an important role in developing another 2D materials and in new nano-optoelectronic devices using it. • Synthesis of 2D material Nb 3 I 8 in centimeter scale. • Experimental estimation of band gap of Nb 3 I 8 about (∼1 eV) and application to IR detectors. • Bulk to few-layers separation of Nb 3 I 8 sheet using mechanical exfoliation method. [ABSTRACT FROM AUTHOR]