ABSTRACT
Nonlinear systems, such as robotic systems, play an increasingly important role in our modern daily life and have become more dominant in many industries; however, robotic control still faces various challenges due to diverse and unstructured work environments. This article proposes a double-loop recurrent neural network (DLRNN) with the support of a Type-2 fuzzy system and a self-organizing mechanism for improved performance in nonlinear dynamic robot control. The proposed network has a double-loop recurrent structure, which enables better dynamic mapping. In addition, the network combines a Type-2 fuzzy system with a double-loop recurrent structure to improve the ability to deal with uncertain environments. To achieve an efficient system response, a self-organizing mechanism is proposed to adaptively adjust the number of layers in a DLRNN. This work integrates the proposed network into a conventional sliding mode control (SMC) system to theoretically and empirically prove its stability. The proposed system is applied to a three-joint robot manipulator, leading to a comparative study that considers several existing control approaches. The experimental results confirm the superiority of the proposed system and its effectiveness and robustness in response to various external system disturbances.
ABSTRACT
Approximation biases of value functions are considered a key problem in reinforcement learning (RL). In particular, existing RL algorithms are hindered by overestimation and underestimation biases, i.e., value mismatching between RL's actual returns and action-value approximations limits the performance of RL algorithms. In this article, we first develop a new synthesis loss function for RL's action-value estimation integrating a regularization term and a modified "clipped double Q -learning" structure for solving overestimation and underestimation biases. To minimize the differences between action-value estimations and actual returns in RL, we develop a new discrepancy function to determine the type and magnitude of approximation biases. Then, two coefficients embedded in the synthesis loss are automatically tuned by minimizing the discrepancy function during training to minimize approximation biases. We further design a new actor-critic (AC) algorithm, named AC with synthesis loss (ACSL), by integrating the synthesis loss function and an error-controlled mechanism. Experimental results on continuous control tasks illustrate that the proposed ACSL algorithm outperforms other cutting-edge RL methods in many tasks and that the proposed synthesis loss function is easily implemented into other algorithms and significantly reduces approximation biases while improving performance. The proposed method can successfully handle many complex continuous control tasks and can greatly outperform other state-of-the-art algorithms on most tasks.
ABSTRACT
Organic molecules have been regarded as ideal candidates for near-infrared (NIR) optoelectronic active materials due to their customizability and ease of large-scale production. However, constrained by the intricate molecular design and severe energy gap law, the realization of optoelectronic devices in the second near-infrared (NIR (II)) region with required narrow band gaps presents more challenges. Herein, we have originally proposed a cocrystal strategy that utilizes intermolecular charge-transfer interaction to drive the redshift of absorption and emission spectra of a series BFXTQ (X = 0, 1, 2, 4) cocrystals, resulting in the spectra located at NIR (II) window and reducing the optical bandgap to â¼0.98 eV. Significantly, these BFXTQ-based optoelectronic devices can exhibit dual-mode optoelectronic characteristics. An investigation of a series of BFXTQ-based photodetectors exhibits detectivity (D*) surpassing 1013 Jones at 375 to 1064 nm with a maximum of 1.76 × 1014 Jones at 1064 nm. Moreover, the radiative transition of CT excitons within the cocrystals triggers NIR emission over 1000 nm with a photoluminescence quantum yield (PLQY) of â¼4.6% as well as optical waveguide behavior with a low optical-loss coefficient of 0.0097 dB/µm at 950 nm. These results promote the advancement of an emerging cocrystal approach in micro/nanoscale NIR multifunctional optoelectronics.
ABSTRACT
The aminated sodium lignosulfonate (AELS) was prepared through a Mannich reaction and characterized via FT-IR, TG, SEM and XPS in this study. Subsequently, the adsorption capacity of AELS for methyl blue (MB) was evaluated under various conditions such as pH, adsorbent dosage, contact time, initial concentration and temperature. The adsorption kinetics, isotherms and thermodynamics of AELS for methyl blue were investigated and analyzed. The results were found to closely adhere to the pseudo-second-order kinetic model and Langmuir isotherm model, suggesting a single-molecular-layer adsorption process. Notably, the maximum adsorption capacity of AELS for methyl blue (153.42 mg g-1) was achieved under the specified conditions (T = 298 K, MAELS = 0.01 g, pH = 6, VMB = 25 mL, C0 = 300 mg L-1). The adsorption process was determined to be spontaneous and endothermic. Following five adsorption cycles, the adsorption capacity exhibited a minimal reduction from 118.99 mg g-1 to 114.33 mg g-1, indicating good stability. This study contributes to the advancement of utilizing natural resources effectively and sustainably.
ABSTRACT
PSNR-oriented models are a critical class of super-resolution models with applications across various fields. However, these models tend to generate over-smoothed images, a problem that has been analyzed previously from the perspectives of models or loss functions, but without taking into account the impact of data properties. In this paper, we present a novel phenomenon that we term the center-oriented optimization (COO) problem, where a model's output converges towards the center point of similar high-resolution images, rather than towards the ground truth. We demonstrate that the strength of this problem is related to the uncertainty of data, which we quantify using entropy. We prove that as the entropy of high-resolution images increases, their center point will move further away from the clean image distribution, and the model will generate over-smoothed images. Implicitly optimizing the COO problem, perceptual-driven approaches such as perceptual loss, model structure optimization, or GAN-based methods can be viewed. We propose an explicit solution to the COO problem, called Detail Enhanced Contrastive Loss (DECLoss). DECLoss utilizes the clustering property of contrastive learning to directly reduce the variance of the potential high-resolution distribution and thereby decrease the entropy. We evaluate DECLoss on multiple super-resolution benchmarks and demonstrate that it improves the perceptual quality of PSNR-oriented models. Moreover, when applied to GAN-based methods, such as RaGAN, DECLoss helps to achieve state-of-the-art performance, such as 0.093 LPIPS with 24.51 PSNR on 4× downsampled Urban100, validating the effectiveness and generalization of our approach.
ABSTRACT
Sleep apnea is a common sleep disorder. Traditional testing and diagnosis heavily rely on the expertise of physicians, as well as analysis and statistical interpretation of extensive sleep testing data, resulting in time-consuming and labor-intensive processes. To address the problems of complex feature extraction, data imbalance, and low model capacity, we proposed an automatic sleep apnea classification model (CA-EfficientNet) based on the wavelet transform, a lightweight neural network, and a coordinated attention mechanism. The signal is converted into a time-frequency image by wavelet transform and put into the proposed model for classification. The effects of input time window, wavelet transform type and data balancing on the classification performance are considered, and a cost-sensitive algorithm is introduced to more accurately distinguish between normal and abnormal breathing events. PhysioNet apnea ECG database was used for training and evaluation. The 3-min Frequency B-Spline wavelets transform of ECG signal was carried out, and Dice Loss was used to train the classification model of sleep breathing. The classification accuracy was 93.44%, sensitivity was 88.9%, specificity was 96.2% and most indexes were better than other related work.
Subject(s)
Deep Learning , Sleep Apnea Syndromes , Sleep Apnea, Obstructive , Humans , Wavelet Analysis , Sleep Apnea, Obstructive/diagnostic imaging , Sleep Apnea Syndromes/diagnostic imaging , Electrocardiography/methodsABSTRACT
Two-dimensional organic lateral heterostructures (2D OLHs) are attractive for the fabrication of functional materials. However, it is difficult to control the nucleation, growth and orientation of two distinct components. Here we report the combination of two methods-liquid-phase growth and vapour-phase growth-to synthesize 2D OLHs from perylene and a perylenecarboxaldehyde derivative, with a lateral size of ~20 µm and a tunable thickness ranging from 20 to 400 nm. The screw dislocation growth behaviour of the 2D crystals shows the spiral arrangement of atoms within the crystal lattice, which avoids volume expansion and contraction of OLH, thereby minimizing lateral connection defects. Selective control of the nucleation and sequential growth of 2D crystals leads to structural inversion of the 2D OLHs by the vapour-phase growth method. The resulting OLHs show good light-transport capabilities and tunable spatial exciton conversion, useful for photonic applications. This synthetic strategy can be extended to other families of organic polycyclic aromatic hydrocarbons, as demonstrated with other pyrene and perylene derivatives.
ABSTRACT
The early detection of liver cancer greatly improves survival rates and allows for less invasive treatment options. As a non-invasive optical detection technique, Surface-Enhanced Raman Spectroscopy (SERS) has shown significant potential in early cancer detection, providing multiple advantages over conventional methods. The majority of existing cancer detection methods utilize multivariate statistical analysis to categorize SERS data. However, these methods are plagued by issues such as information loss during dimensionality reduction and inadequate ability to handle nonlinear relationships within the data. To overcome these problems, we first use wavelet transform with its multi-scale analysis capability to extract multi-scale features from SERS data while minimizing information loss compared to traditional methods. Moreover, deep learning is employed for classification, leveraging its strong nonlinear processing capability to enhance accuracy. In addition, the chosen neural network incorporates a data augmentation method, thereby enriching our training dataset and mitigating the risk of overfitting. Moreover, we acknowledge the significance of selecting the appropriate wavelet basis functions in SERS data processing, prompting us to choose six specific ones for comparison. We employ SERS data from serum samples obtained from both liver cancer patients and healthy volunteers to train and test our classification model, enabling us to assess its performance. Our experimental results demonstrate that our method achieved outstanding and healthy volunteers to train and test our classification model, enabling us to assess its performance. Our experimental results demonstrate that our method achieved outstanding performance, surpassing the majority of multivariate statistical analysis and traditional machine learning classification methods, with an accuracy of 99.38 %, a sensitivity of 99.8 %, and a specificity of 97.0 %. These results indicate that the combination of SERS, wavelet transform, and deep learning has the potential to function as a non-invasive tool for the rapid detection of liver cancer.
Subject(s)
Deep Learning , Liver Neoplasms , Humans , Spectrum Analysis, Raman/methods , Multivariate Analysis , Neural Networks, Computer , Liver Neoplasms/diagnosisABSTRACT
Introduction: Internal fistula across the posterior wall of stomach and the transverse colon caused by foreign bodies in the alimentary tract presents an extremely rare medical entity. Presentation of case: We report an aschizophrenia female patient with onset of internal fistula across the posterior wall of stomach and the transverse colon triggered by swallowed magnetic metal beads. The patient was admitted to the emergency room of Northern Jiangsu People's Hospital because of acute right lower abdominal pain. Emergency routine abdominal CT scan revealed acute appendicitis and a set of foreign body in digestive tract. Discussion: The foreign body in the stomach was removed by open surgery after tentative Endoscopic foreign body removal and laparoscopic appendectomy and exploration. In the process of exploring the gastric wall, it was found that one of magnet beads was embedded in the posterior wall of stomach and adhered to part of the transverse colon. After separation, it was found that an internal fistula was formed across the posterior wall of stomach and the transverse colon. As the patient ate only a small amount of food within 2 days, and the intestines were in good condition, we performed partial transverse colectomy, end-to-side anastomosis and gastric wall repair. Conclusion: This case shows that for long-term foreign bodies in the digestive tract, we should be beware of the onset of gastrointestinal perforation. Moreover, perforation caused by the force acting on a blunt foreign body often results in atypical imaging findings, and the diagnosis of perforation cannot be clearly determined by imaging findings such as the presence of free gas downstream of the diaphragm. This poses new challenges for clear diagnosis and treatment.
ABSTRACT
Cellulose-based carbon (CBC) is widely known for its porous structure and high specific surface area and is liable to adsorb gas molecules and macromolecular pollutants. However, the application of CBC in gas sensing has been little studied. In this paper, a ZnO/CBC heterojunction was formed by means of simple co-precipitation and high-temperature carbonization. As a new ammonia sensor, the prepared ZnO/CBC sensor can detect ammonia that the previous pure ZnO ammonia sensor cannot at room temperature. It has a great gas sensing response, stability, and selectivity to an ammonia concentration of 200 ppm. This study provides a new idea for the design and synthesis of biomass carbon-metal oxide composites.
ABSTRACT
The aim of the study was to investigate the biomechanical behavior of three-dimensionally (3D)-printed surgical plates used for mandibular defect reconstruction, compare them with conventional surgical plates, and provide experimental evidence for their clinical application. Three-dimensional models were created for the normal mandible and for mandibular body defects reconstructed using free fibula and deep circumflex iliac artery flaps. Three-dimensional finite element models of reconstructed mandibles fixed using 3D-printed and conventional surgical plates were established. Vertical occlusal forces were applied to the remaining teeth and the displacement and Von Mises stress distributions were studied using finite element analysis. The normal and reconstructed mandibles had similar biomechanical behaviors. The displacement distributions for the surgical plates were similar, and the maximum total deformation occurred at the screw hole of the anterior segment of the surgical plates. However, there were differences in the Von Mises stress distributions for the surgical plates. In reconstructed mandibles fixed using 3D-printed surgical plates, the maximum equivalent Von Mises stress occurred at the screw hole of the posterior segment, while in those fixed using conventional surgical plates, the maximum equivalent Von Mises stress was at the screw hole of the anterior segment. In the mandible models reconstructed with the same free flap but fixed with different surgical plates, the plates had similar biomechanical behaviors. The biomechanical behavior of 3D-printed surgical plates was similar to conventional surgical plates, suggesting that 3D-printed surgical plates used to reconstruct mandibular body defects with vascularized autogenous bone grafts could lead to secure and stable fixation.
Subject(s)
Bone Plates , Mandible , Humans , Finite Element Analysis , Mandible/surgery , Bone Screws , Printing, Three-DimensionalABSTRACT
Two metal chelates of Dioscorea oppositifolia L. peel polysaccharides (DTP) were prepared: iron chelate (DTP-Fe) and zinc chelate (DTP-Zn). The physicochemical properties of the polysaccharide and its metal chelates were assessed by UV-Vis absorption spectroscopy, Fourier-transform infra-red spectroscopy, scanning electron microscopy, and thermogravimetric analysis. Antioxidant activities were evaluated by DPPH, ABTS + and hydroxyl radical scavenging assays. According to ICP-MS, the iron content of DTP-Fe was 9.47%, while the zinc content of DTP-Zn was 4.02%. The antioxidant capacity of DTP-Fe increased with the increase of concentration, and its overall activity was higher than that of DTP and DTP-Zn. This polysaccharide-iron chelate can be developed and utilised as an antioxidant and multifunctional iron supplement. DTP-Zn showed the potential to be a natural antioxidant and zinc supplement food.
ABSTRACT
Network pruning is an effective approach to reduce network complexity with acceptable performance compromise. Existing studies achieve the sparsity of neural networks via time-consuming weight training or complex searching on networks with expanded width, which greatly limits the applications of network pruning. In this paper, we show that high-performing and sparse sub-networks without the involvement of weight training, termed "lottery jackpots", exist in pre-trained models with unexpanded width. Our presented lottery jackpots are traceable through empirical and theoretical outcomes. For example, we obtain a lottery jackpot that has only 10% parameters and still reaches the performance of the original dense VGGNet-19 without any modifications on the pre-trained weights on CIFAR-10. Furthermore, we improve the efficiency for searching lottery jackpots from two perspectives. First, we observe that the sparse masks derived from many existing pruning criteria have a high overlap with the searched mask of our lottery jackpot, among which, the magnitude-based pruning results in the most similar mask with ours. In compliance with this insight, we initialize our sparse mask using the magnitude-based pruning, resulting in at least 3× cost reduction on the lottery jackpot searching while achieving comparable or even better performance. Second, we conduct an in-depth analysis of the searching process for lottery jackpots. Our theoretical result suggests that the decrease in training loss during weight searching can be disturbed by the dependency between weights in modern networks. To mitigate this, we propose a novel short restriction method to restrict change of masks that may have potential negative impacts on the training loss, which leads to a faster convergence and reduced oscillation for searching lottery jackpots. Consequently, our searched lottery jackpot removes 90% weights in ResNet-50, while it easily obtains more than 70% top-1 accuracy using only 5 searching epochs on ImageNet.
ABSTRACT
Organic hierarchical branch micro/nanostructures constituted by single crystals with inherent multichannel characteristics exhibit superior potential in regulating photon transmission for photonic circuits. However, organic branch micro/nanostructures with precise branch positions are extremely difficult to achieve due to the randomness of the nucleation process. Herein, by taking advantage of the dislocation stress field-impurity interaction that solute molecules deposit preferentially along the dislocation line, twinning deformation was introduced into microcrystals to induce oriented nucleation sites, and ultimately organic branch microstructures with controllable branch sites were fabricated. The growth mechanism of these controllable single crystals with an angle of 140° between trunk and branch is attributed to the low lattice mismatching ratio (η) of 4.8%. These as-prepared hierarchical branch single crystals with asymmetrical optical waveguide characteristics have been demonstrated as an optical logic gate with multiple input/out channels, which provides a route to command the nucleation sites and offers potential applications in the organic optoelectronics at the micro/nanoscale.
ABSTRACT
The mainstream approach for filter pruning is usually either to force a hard-coded importance estimation upon a computation-heavy pretrained model to select "important" filters, or to impose a hyperparameter-sensitive sparse constraint on the loss objective to regularize the network training. In this paper, we present a novel filter pruning method, dubbed dynamic-coded filter fusion (DCFF), to derive compact CNNs in a computation-economical and regularization-free manner for efficient image classification. Each filter in our DCFF is first given an inter-similarity distribution with a temperature parameter as a filter proxy, on top of which, a fresh Kullback-Leibler divergence based dynamic-coded criterion is proposed to evaluate the filter importance. In contrast to simply keeping high-score filters in other methods, we propose the concept of filter fusion, i.e., the weighted averages using the assigned proxies, as our preserved filters. We obtain a one-hot inter-similarity distribution as the temperature parameter approaches infinity. Thus, the relative importance of each filter can vary along with the training of the compact CNN, leading to dynamically changeable fused filters without both the dependency on the pretrained model and the introduction of sparse constraints. Extensive experiments on classification benchmarks demonstrate the superiority of our DCFF over the compared counterparts. For example, our DCFF derives a compact VGGNet-16 with only 72.77M FLOPs and 1.06M parameters while reaching top-1 accuracy of 93.47% on CIFAR-10. A compact ResNet-50 is obtained with 63.8% FLOPs and 58.6% parameter reductions, retaining 75.60% top-1 accuracy on ILSVRC-2012. Our code, narrower models and training logs are available at https://github.com/lmbxmu/DCFF.
ABSTRACT
When a different types of reactor are operating at the same area and the same period of time, released radionuclides are hard to follow in the environment. In general, isotopic techniques can be used for source localization. To obtain the distribution of hydrogen isotope in soil, eight sampling points were selected along the local dominant wind direction with different distances away from Qinshan Nuclear Power Plant, and soil samples at different depths (0-2, 2-5, 5-10, 10-20, 20-30 cm) were collected in December 2019 and December 2020, respectively. The concentrations of hydrogen isotopes were measured in the soil samples at different depth. The spatial distribution of tritium and deuterium in the surface soil was related to soil properties and the distance from the nuclear power plant. It was found that tritium and deuterium are generally enriched in the upper layer. Determination of the deuterium concentration in the environment may be a new way to trace the released tritium from the reactors.
Subject(s)
Radiation Monitoring , Soil , Nuclear Power Plants , Hydrogen , Tritium/analysis , Deuterium , Radiation Monitoring/methodsABSTRACT
Organic charge transfer (CT) cocrystals open a new door for the exploitation of low-dimensional near-infrared (NIR) emitters by a convenient self-assembly approach. However, research about the fabrication of sheet-like NIR-emitting microstructures that are significant for structural construction and integrated application is limited by the unidirectional molecular packing mode. Herein, via regulation of the biaxial intermolecular CT interaction, single-crystalline microsheets with remarkable NIR emission from 720 to 960 nm were synthesized via the solution self-assembly process of dithieno[3,2-b:2',3'-d]thiophene and 7,7,8,8-tetracyanoquinodimethane. The expected sheet-like structure is conducive to achieving a two-dimensional (2D) optical waveguide with an ultralow optical loss rate of 0.250 dB/µm at 860 nm. More significantly, these as-prepared organic microsheets with tunable thicknesses (h) from 100 to 1100 nm exhibit thickness-dependent NIR optical transportation performance. These findings could pave the way to a new class of low-dimensional NIR emitters for 2D photonics at telecom wavelengths.
ABSTRACT
High fructose intake is an essential risk factor for kidney injury. However, the specific mechanism underlying high fructose-induced kidney injury remains unclarified. Carbohydrate response element-binding protein (ChREBP) is a key transcriptional activator that regulates fructose metabolism. ChREBP-ß exhibits sustained activity due to the lack of a low glucose inhibitory domain, and is thus described as the active form of ChREBP. In this study, a mouse model with specific overexpression of ChREBP-ß in the renal tubule was established by using the Cre/LoxP method. Quantitative proteomic analysis and experimental verification results suggest that ChREP-ß overexpression leads to ferroptosis of renal tubular epithelial cells and kidney injury. ChREPB-ß promotes the gene transcription of thioredoxin-interacting protein (TXNIP) and thereby increases its expression level. TXNIP is associated with activation of ferroptosis. TXNIP can initiate ferroptosis and eventually contribute to high fructose-induced renal tubular epithelial cell damage. Through down-regulating ChREBP-ß, metformin can inhibit gene transcription of TXNIP, attenuate high fructose-induced ferroptosis in renal tubular epithelial cells, and alleviate kidney injury. In conclusion, ChREBP-ß mediates fructose-induced ferroptosis of renal tubular epithelial cells, and metformin with a ChREBP-ß inhibitory effect may be a potential treatment for ferroptosis of renal tubular epithelial cells.
Subject(s)
Ferroptosis , Metformin , Mice , Animals , Ferroptosis/genetics , Proteomics , Glucose/metabolism , Epithelial Cells/metabolism , Metformin/pharmacology , Kidney Tubules/metabolism , Fructose , Basic Helix-Loop-Helix Leucine Zipper Transcription Factors/genetics , Basic Helix-Loop-Helix Leucine Zipper Transcription Factors/metabolism , Carrier Proteins/genetics , Thioredoxins/genetics , Thioredoxins/metabolismABSTRACT
Though network pruning receives popularity in reducing the complexity of convolutional neural networks (CNNs), it remains an open issue to concurrently maintain model accuracy as well as achieve significant speedups on general CPUs. In this paper, we propose a novel 1×N pruning pattern to break this limitation. In particular, consecutive N output kernels with the same input channel index are grouped into one block, which serves as a basic pruning granularity of our pruning pattern. Our 1×N pattern prunes these blocks considered unimportant. We also provide a workflow of filter rearrangement that first rearranges the weight matrix in the output channel dimension to derive more influential blocks for accuracy improvements and then applies similar rearrangement to the next-layer weights in the input channel dimension to ensure correct convolutional operations. Moreover, the output computation after our 1×N pruning can be realized via a parallelized block-wise vectorized operation, leading to significant speedups on general CPUs. The efficacy of our pruning pattern is proved with experiments on ILSVRC-2012. For example, given the pruning rate of 50% and N=4, our pattern obtains about 3.0% improvements over filter pruning in the top-1 accuracy of MobileNet-V2. Meanwhile, it obtains 56.04ms inference savings on Cortex-A7 CPU over weight pruning. Our project is made available at https://github.com/lmbxmu/1xN.
ABSTRACT
Binary neural networks (BNNs) have attracted broad research interest due to their efficient storage and computational ability. Nevertheless, a significant challenge of BNNs lies in handling discrete constraints while ensuring bit entropy maximization, which typically makes their weight optimization very difficult. Existing methods relax the learning using the sign function, which simply encodes positive weights into +1s, and -1s otherwise. Alternatively, we formulate an angle alignment objective to constrain the weight binarization to {0,+1} to solve the challenge. In this article, we show that our weight binarization provides an analytical solution by encoding high-magnitude weights into +1s, and 0s otherwise. Therefore, a high-quality discrete solution is established in a computationally efficient manner without the sign function. We prove that the learned weights of binarized networks roughly follow a Laplacian distribution that does not allow entropy maximization, and further demonstrate that it can be effectively solved by simply removing the l2 regularization during network training. Our method, dubbed sign-to-magnitude network binarization (SiMaN), is evaluated on CIFAR-10 and ImageNet, demonstrating its superiority over the sign-based state-of-the-arts. Our source code, experimental settings, training logs and binary models are available at https://github.com/lmbxmu/SiMaN.