Search

The linear classifier is widely used in various image classification tasks. It works by optimizing the distance between a sample and its corresponding class center. However, in real-world data, one class can contain several local clusters, e.g., birds of different poses. To address this complexity, we propose a novel multi-center classifier. Different from the vanilla linear classifier, our proposal is established on the assumption that the deep features of the training set follow a Gaussian Mixture distribution. Specifically, we create a conditional Gaussian distribution for each class and then sample multiple sub-centers from that distribution to extend the linear classifier. This approach allows the model to capture intra-class local structures more efficiently. In addition, at test time we set the mean of the conditional Gaussian distribution as the class center of the linear classifier and follow the vanilla linear classifier outputs, thus requiring no additional parameters or computational overhead. Extensive experiments on image classification show that the proposed multi-center classifier is a powerful alternative to widely used linear classifiers. Code available at https://github.com/ZheminZhang1/MultiCenter-Classifier.

Hui Ke,* Xuwei Gui,* Wenwen Sun, Shaojun Zhang, Yan Yang, Zhemin Zhang, Lin Fan Department of Tuberculosis, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai Clinic and Research Center of Tuberculosis, Shanghai Key Laboratory of Tuberculosis, Shanghai, People’s Republic of China*These authors contributed equally to this workCorrespondence: Lin Fan; Zhemin Zhang, Email fanlinsj@163.com; zhemindoc@163.comObjective: To evaluate the safety, tolerability, and efficacy of prolonged bedaquiline (Bdq) treatment in patients with multi-drug/rifampin-resistant tuberculosis (MDR/RR-TB).Methods: This prospective cohort study was performed from August 2018 to August 2021. Patients diagnosed with MDR/RR-TB who met the inclusion criteria were prospectively included. Patients were treated with individual regimens of 18– 20 months containing Bdq for six months or a prolonged course of nine or 12 months according to treatment demands, and the efficacy and safety with a different course of Bdq-containing regimens were compared and evaluated.Results: A total of 159 MDR/RR-TB patients were included in the study, including 96 cases with six months of Bdq, 50 cases with nine months of Bdq, and 13 patients with 12 months of Bdq. The treatment success rates were 89.6%, 90%, and 84.6% in Bdq at six months, nine months, and 12 months, respectively, which were not statistically different (P = 0.85). The main adverse events (AEs) were anemia, thrombocytopenia, and liver dysfunction in all patients, with no significant difference among the three groups. Patients who had fewer drugs chosen, disseminated lesions or lesions that were slowly absorbed, and severe cavities were the common reasons for prolonged use of Bdq.Conclusion: Prolonged course use of Bdq from six months to 12 months clinically proved to be safe and efficient, and patients with severe or disseminated lesions had the chance to prolong the use of Bdq for more than six months to a

Jinpeng Shi,1,* Jiayu Li,1,* Haowei Wang,1,* Xuefei Li,2 Qi Wang,1 Chao Zhao,2 Lei Cheng,2 Ruoshuang Han,1 Peixin Chen,1 Haoyue Guo,1 Zhuoran Tang,1 Caicun Zhou,1 Zhemin Zhang,1 Fengying Wu1 1Department of Medical Oncology, Shanghai Pulmonary Hospital & Thoracic Cancer Institute, School of Medicine, Tongji University, Shanghai, People’s Republic of China; 2Department of Lung Cancer and Immunology, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, People’s Republic of China*These authors contributed equally to this workCorrespondence: Fengying Wu; Zhemin Zhang, Department of Medical Oncology, Shanghai Pulmonary Hospital & Thoracic Cancer Institute, School of Medicine, Tongji University, No. 507, Zheng Min Road, Shanghai, 200433, People’s Republic of China, Tel +86 21 6511 5006, Email fywu@163.com; zhemindoc@163.comPurpose: Neutrophils act as a non-negligible regulator in the initiation and progression of malignancies, playing bifacial roles in the process. Thus, to understand the heterogeneity of tumor-associated neutrophils (TANs) comprehensively in advanced non-small cell lung cancer (NSCLC) at single-cell resolution is necessary and urgent.Materials and Methods: We applied single-cell RNA-sequencing (scRNA-seq) to portray the subtype-specific transcriptome landscape of TANs in advanced NSCLC using nine freshly obtained specimens. The scRNA-seq data were further processed for pseudo-time analysis to depict the developmental trajectory of TANs. Meanwhile, the interplay between TANs and other cell types within tumor microenvironment (TME) was revealed by intercellular interaction analysis.Results: Seven distinct TAN subtypes were defined, of which, the N3 cluster was considered inflammatory phenotype expressing genes encoding multiple chemotactic cytokines, and correlated with inferior overall survival, indicating that N3 might be a pro-tumorigenic TAN subtype. N1 and N5 clusters were considere

Recently, Transformers have shown promising performance in various vision tasks. However, the high costs of global self-attention remain challenging for Transformers, especially for high-resolution vision tasks. Local self-attention runs attention computation within a limited region for the sake of efficiency, resulting in insufficient context modeling as their receptive fields are small. In this work, we introduce two new attention modules to enhance the global modeling capability of the hierarchical vision transformer, namely, random sampling windows (RS-Win) and important region windows (IR-Win). Specifically, RS-Win sample random image patches to compose the window, following a uniform distribution, i.e., the patches in RS-Win can come from any position in the image. IR-Win composes the window according to the weights of the image patches in the attention map. Notably, RS-Win is able to capture global information throughout the entire model, even in earlier, high-resolution stages. IR-Win enables the self-attention module to focus on important regions of the image and capture more informative features. Incorporated with these designs, RSIR-Win Transformer demonstrates competitive performance on common vision tasks., Comment: arXiv admin note: substantial text overlap with arXiv:2209.08726

Recently, Transformers have shown promising performance in various vision tasks. However, the high costs of global self-attention remain challenging for Transformers, especially for high-resolution vision tasks. Inspired by one of the most successful transformers-based models for NLP: Big Bird, we propose a novel sparse attention mechanism for Vision Transformers (ViT). Specifically, we separate the heads into three groups, the first group used convolutional neural network (CNN) to extract local features and provide positional information for the model, the second group used Random Sampling Windows (RS-Win) for sparse self-attention calculation, and the third group reduces the resolution of the keys and values by average pooling for global attention. Based on these components, ViT maintains the sparsity of self-attention while maintaining the merits of Big Bird (i.e., the model is a universal approximator of sequence functions and is Turing complete). Moreover, our results show that the positional encoding, a crucial component in ViTs, can be safely removed in our model. Experiments show that Vision Big Bird demonstrates competitive performance on common vision tasks., Comment: arXiv admin note: substantial text overlap with arXiv:2304.06250

Recently, Transformers have shown promising performance in various vision tasks. A challenging issue in Transformer design is that global self-attention is very expensive to compute, especially for the high-resolution vision tasks. Local self-attention performs attention computation within a local region to improve its efficiency, which leads to their receptive fields in a single attention layer are not large enough, resulting in insufficient context modeling. When observing a scene, humans usually focus on a local region while attending to non-attentional regions at coarse granularity. Based on this observation, we develop the axially expanded window self-attention mechanism that performs fine-grained self-attention within the local window and coarse-grained self-attention in the horizontal and vertical axes, and thus can effectively capturing both short- and long-range visual dependencies.

Attention mechanism, being frequently used to train networks for better feature representations, can effectively disentangle the target object from irrelevant objects in the background. Given an arbitrary image, we find that the background's irrelevant objects are most likely to occlude/block the target object. We propose, based on this finding, a ReplaceBlock to simulate the situations when the target object is partially occluded by the objects that are deemed as background. Specifically, ReplaceBlock erases the target object in the image, and then generates a feature map with only irrelevant objects and background by the model. Finally, some regions in the background feature map are used to replace some regions of the target object in the original image feature map. In this way, ReplaceBlock can effectively simulate the feature map of the occluded image. The experimental results show that ReplaceBlock works better than DropBlock in regularizing convolutional networks.

Treating class with a single center may hardly capture data distribution complexities. Using multiple sub-centers is an alternative way to address this problem. However, highly correlated sub-classes, the classifier's parameters grow linearly with the number of classes, and lack of intra-class compactness are three typical issues that need to be addressed in existing multi-subclass methods. To this end, we propose to use Fixed Sub-Center (F-SC), which allows the model to create more discrepant sub-centers while saving memory and cutting computational costs considerably. The F-SC specifically, first samples a class center Ui for each class from a uniform distribution, and then generates a normal distribution for each class, where the mean is equal to Ui. Finally, the sub-centers are sampled based on the normal distribution corresponding to each class, and the sub-centers are fixed during the training process avoiding the overhead of gradient calculation. Moreover, F-SC penalizes the Euclidean distance between the samples and their corresponding sub-centers, it helps remain intra-compactness. The experimental results show that F-SC significantly improves the accuracy of both image classification and fine-grained recognition tasks.

We propose Self-Supervised Implicit Attention (SSIA), a new approach that adaptively guides deep neural network models to gain attention by exploiting the properties of the models themselves. SSIA is a novel attention mechanism that does not require any extra parameters, computation, or memory access costs during inference, which is in contrast to existing attention mechanism. In short, by considering attention weights as higher-level semantic information, we reconsidered the implementation of existing attention mechanisms and further propose generating supervisory signals from higher network layers to guide lower network layers for parameter updates. We achieved this by building a self-supervised learning task using the hierarchical features of the network itself, which only works at the training stage. To verify the effectiveness of SSIA, we performed a particular implementation (called an SSIA block) in convolutional neural network models and validated it on several image classification datasets. The experimental results show that an SSIA block can significantly improve the model performance, even outperforms many popular attention methods that require additional parameters and computation costs, such as Squeeze-and-Excitation and Convolutional Block Attention Module. Our implementation will be available on GitHub.

Positional encoding is important for vision transformer (ViT) to capture the spatial structure of the input image. General effectiveness has been proven in ViT. In our work we propose to train ViT to recognize the positional label of patches of the input image, this apparently simple task actually yields a meaningful self-supervisory task. Based on previous work on ViT positional encoding, we propose two positional labels dedicated to 2D images including absolute position and relative position. Our positional labels can be easily plugged into various current ViT variants. It can work in two ways: (a) As an auxiliary training target for vanilla ViT (e.g., ViT-B and Swin-B) for better performance. (b) Combine the self-supervised ViT (e.g., MAE) to provide a more powerful self-supervised signal for semantic feature learning. Experiments demonstrate that with the proposed self-supervised methods, ViT-B and Swin-B gain improvements of 1.20% (top-1 Acc) and 0.74% (top-1 Acc) on ImageNet, respectively, and 6.15% and 1.14% improvement on Mini-ImageNet.

Siyuan He,1,2,* Yuzhen Zou,1,2,* Mengling Zhan,1,2,* Qi Guo,1,2 Yongjie Zhang,1,2 Zhemin Zhang,1 Bing Li,1 Shaoyan Zhang,1 Haiqing Chu1,3 1Department of Respiratory Medicine, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai 200433, People’s Republic of China; 2Tongji University School of Medicine, Shanghai 200092, People’s Republic of China; 3Shanghai Key Laboratory of Tuberculosis, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai 200433, People’s Republic of China*These authors contributed equally to this workCorrespondence: Haiqing Chu; Shaoyan ZhangDepartment of Respiratory Medicine, Shanghai Pulmonary Hospital, Tongji University School of Medicine, No. 507 Zhengmin Road, Shanghai 200433, People’s Republic of ChinaTel/ Fax +86 21 6511 5006Email chu_haiqing@126.com; zhangshaoyan000@163.comPurpose: Imipenem is one of the very few effective options for treating Mycobacterium abscessus (M. abscessus) infections; the development of imipenem resistance is a major health concern.Materials and Methods: The susceptibility of 194 clinical M. abscessus isolates to imipenem was determined. The ability of imipenem to synergize with N,N,N′,N′-tetrakis(2-pyridylmethyl)ethylenediamine (TPEN), a zinc chelator and a metallo-β-lactamases (MBLs) inhibitor, to inhibit M. abscessus growth was also assessed.Results: M. abscessus exhibited an elevated resistance to imipenem (MIC50 = 16 mg/L, MIC90 = 64 mg/L). A combination of TPEN and imipenem synergized to inhibit the growth of 100% of imipenem-resistant and 79.2% of imipenem-resistance intermediate isolates; no synergy was observed treating imipenem-sensitive isolates. A remarkable decrease in the MIC50 (from 16 to 4 mg/L) and MIC90 (from 64 to 8 mg/L) of imipenem was observed when it was combined with TPEN; the portion of imipenem-resistant isolates also decreased (from 48.4% to 0%). Consistent with these results

Qi Guo,1,2 Jianhui Chen,1,2 Shaoyan Zhang,1 Yuzhen Zou,1,2 Yongjie Zhang,1,2 Dongdong Huang,1 Zhemin Zhang,1 Bing Li,1 Haiqing Chu1,3 1Department of Respiratory Medicine, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai 200433, People’s Republic of China; 2Tongji University School of Medicine, Shanghai 200092, People’s Republic of China; 3Shanghai Key Laboratory of Tuberculosis, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai 200433, People’s Republic of ChinaCorrespondence: Haiqing Chu; Bing LiDepartment of Respiratory Medicine, Shanghai Pulmonary Hospital, Tongji University School of Medicine, No. 507 Zhengmin Road, Shanghai 200433, People’s Republic of ChinaTel/Fax +86 021 6511 5006Email chu_haiqing@126.com; libing044162@163.comPurpose: The emergence of clarithromycin resistance is a challenge in treating Mycobacterium abscessus infections. Known mechanisms that contribute to intrinsic clarithromycin resistance focus on rrl gene-related mutations, but resistant clinical isolates often exhibit an inconsistent rrl genotype.Patients and Methods: In this study, 194 clinical Mycobacterium abscessus isolates were collected from patients with lung infections and the whole genome of each isolate was sequenced. A comprehensive examination of the molecular mechanisms underlying intrinsic clarithromycin resistance was performed, combining MIC determination, comparative genome sequence analysis and qRT-PCR.Results: Of the 194 isolates, 13 (6.7%) were clarithromycin resistant; only seven of these harbored a rrl 2270/2271 mutation. The remaining six resistant isolates did not exhibit a specific resistance-associated mutation in the clarithromycin target-site genes, rrl, rplC, rplD and rplV, or in the rrl modification gene erm(41). qRT-PCR analysis showed that the increased expression of the efflux pump genes, MAB_2355c, MAB_1409c and MAB_1846, as well as their positive regulatory gen

Motivated by the recent commercialization of gallium-nitride (GaN) switches, an effort was initiated to determine whether it was feasible to switch the flyback converter at 5 MHz in order to improve the power density of this versatile isolated topology. Soft switching techniques have to be utilized to eliminate the switching loss to maintain high efficiency at multi-megahertz. Compared to the traditional modeling of zero-voltage-switching quasi-square-wave converters, a numerical methodology of parameters design is proposed based on the steady-state model of zero-voltage switching quasi-square-wave flyback converter. The magnetizing inductance is selected to guarantee zero-voltage switching for the entire input and load range with the trade-off design for conduction loss and turn-off loss. A design methodology is introduced to select a minimum core volume for an inductor or coupled inductors experiencing appreciable core loss. The geometric constant Kgac = MLT/(Ac2WA) is shown to be a power function of the core volume Ve, where Ac is the effective core area, WA is the area of the winding window, and MLT is the mean length per turn for commercial toroidal, ER, and PQ cores, permitting the total loss to be expressed as a direct function of the core volume. The inductor is designed to meet specific loss or thermal constraints. An iterative procedure is described in which two- or three-dimensional proximity effects are first neglected and then subsequently incorporated via finite-element simulation. Interleaved and non-interleaved planar PCB winding structures were also evaluated to minimize leakage inductance, self-capacitance and winding loss. The analysis on the trade-off between magnetic size, frequency, loss and temperature indicated the potential for a higher density flyback converter. A small-signal equivalent circuit of QSW converter was proposed to design the control loop and to understand the small-signal behavior. By adding a simple damping resistor on the tr