UVaT: Uncertainty Incorporated View-Aware Transformer for Robust Multi-View Classification.

Existing multi-view classification algorithms usually assume that all examples have observations on all views, and the data in different views are clean. However, in real-world applications, we are often provided with data that have missing representations or contain noise on some views (i.e., missing or noise views). This may lead to significant performance degeneration, and thus many algorithms are proposed to address the incomplete view or noisy view issues. However, most of existing algorithms deal with the two issues separately, and hence may fail when both missing and noisy views exist. They are also usually not flexible in that the view or feature significance cannot be adaptively identified. Besides, the view missing patterns may vary in the training and test phases, and such difference is often ignored. To remedy these drawbacks, we propose a novel multi-view classification framework that is simultaneously robust to both incomplete and noisy views. This is achieved by integrating early fusion and late fusion in a single framework. Specifically, in our early fusion module, we propose a view-aware transformer to mask the missing views and adaptively explore the relationships between views and target tasks to deal with missing views. Considering that view missing patterns may change from the training to the test phase, we also design single-view classification and category-consistency constraints to reduce the dependence of our model on view-missing patterns. In our late fusion module, we quantify the view uncertainty in an ensemble way to estimate the noise level of that view. Then the uncertainty and prediction logits of different views are integrated to make our model robust to noisy views. The framework is trained in an end-to-end manner. Experimental results on diverse datasets demonstrate the robustness and effectiveness of our model for both incomplete and noisy views. Codes are available at https://github.com/li-yapeng/UVaT.

Pseudo-Labeling Based Practical Semi-Supervised Meta-Training for Few-Shot Learning.

Most existing few-shot learning (FSL) methods require a large amount of labeled data in meta-training, which is a major limit. To reduce the requirement of labels, a semi-supervised meta-training (SSMT) setting has been proposed for FSL, which includes only a few labeled samples and numbers of unlabeled samples in base classes. However, existing methods under this setting require class-aware sample selection from the unlabeled set, which violates the assumption of unlabeled set. In this paper, we propose a practical semi-supervised meta-training setting with truly unlabeled data to facilitate the applications of FSL in realistic scenarios. To better utilize both the labeled and truly unlabeled data, we propose a simple and effective meta-training framework, called pseudo-labeling based meta-learning (PLML). Firstly, we train a classifier via common semi-supervised learning (SSL) and use it to obtain the pseudo-labels of unlabeled data. Then we build few-shot tasks from labeled and pseudo-labeled data and design a novel finetuning method with feature smoothing and noise suppression to better learn the FSL model from noise labels. Surprisingly, through extensive experiments across two FSL datasets, we find that this simple meta-training framework effectively prevents the performance degradation of various FSL models under limited labeled data, and also significantly outperforms the representative SSMT models. Besides, benefiting from meta-training, our method also improves several representative SSL algorithms as well. We provide the training code and usage examples at https://github.com/ouyangtianran/PLML.

Identification of novel CDH23 variants linked to hearing loss in a Chinese family: A case report.

RATIONALE: Deafness is associated with both environmental and genetic factors, with hereditary deafness often caused by mutations in deafness-related genes. Identifying and analyzing deafness-related genes will aid in early diagnosis and pave the way for treating inherited deafness through gene therapy in the future. PATIENT CONCERNS: A 15-month-old girl underwent audiological examination at the outpatient clinic of the hospital due to hearing loss and her brother was diagnosed with profound bilateral sensorineural hearing loss at the age of 3. DIAGNOSES: The diagnosis was determined as extremely severe sensorineural hearing loss caused by genetic factors. INTERVENTIONS: Clinical data of the patient were collected, and peripheral blood samples were obtained from both the patient and her family members for DNA extraction and sequencing. OUTCOMES: By utilizing targeted capture next-generation sequencing to further screen for deafness-related genes, 2 novel variants in CDH23 were identified as the causative factors for the patient's deafness. LESSONS: This study identified 2 novel heterozygous mutations in a Chinese family. Both the proband and her sibling have non-syndromic hearing loss (NSHL) and carry distinct heterozygous mutations of cadherin-like 23 (CDH23). One mutation, CDH23:c.2651 A>G, originated from their mother and paternal family, affecting the exon23 domain of CDH23. The other mutation, CDH23:c.2113 G>T, was inherited from their paternal grandmother, impacting the exon19 domain of CDH23. These 2 novel mutations likely cause NSHL by affecting protein function. This finding suggests that identifying 2 novel mutations in CDH23 contributes to the genetic basis of NSHL.

MixIR: Mixing Input and Representations for Contrastive Learning.

Recently, contrastive learning has shown significant progress in learning visual representations from unlabeled data. The core idea is training the backbone to be invariant to different augmentations of an instance. While most methods only maximize the feature similarity between two augmented data, we further generate more challenging training samples and force the model to keep predicting aggregated representation on these hard samples. In this article, we propose MixIR, a mixture-based approach upon the traditional Siamese network. On the one hand, we input two augmented images of an instance to the backbone and obtain the aggregated representation by performing an elementwise maximum of two features. On the other hand, we take the mixture of these augmented images as input and expect the model prediction to be close to the aggregated representation. In this way, the model could access more variant data samples of an instance and keep predicting invariant representations for them. Thus, the learned model is more discriminative compared with previous contrastive learning methods. Extensive experiments on large-scale datasets show that MixIR steadily improves the baseline and achieves competitive results with state-of-the-art methods. Our code is available at https://github.com/happytianhao/MixIR.

Catalyst-free reactions of anilines with ß-chloroenones: synthesis of α-chloroenaminones and 1,4-benzodiazepines.

The Michael addition of anilines to ß-chloroenones gives enaminones by the elimination of hydrochloric acid (HCl). These enaminones are transformed into α-chloroenaminones via in situ sp2 C-H functionalization. Anilines that are attached to an electron-donating group react more readily with ß-chloroenone to give the corresponding products in excellent yields. A highly atom-economical method has been developed using dimethyl sulfoxide (DMSO) as a green oxidant and solvent. The desired α-functionalized enaminones are formed in good yields with excellent Z-selectivity. We have established the generality of this reaction with many substrates, and scaled-up reactions have been performed to showcase the practical applications. A catalyst-free double annulation of ß-chloroenones with o-phenylenediamine has also been demonstrated for the synthesis of 1,4-benzodiazepine derivatives in moderate yields under mild reaction conditions.

Fast Projected Fuzzy Clustering With Anchor Guidance for Multimodal Remote Sensing Imagery.

Multimodal remote sensing image recognition is a popular research topic in the field of remote sensing. This recognition task is mostly solved by supervised learning methods that heavily rely on manually labeled data. When the labels are absent, the recognition is challenging for the large data size, complex land-cover distribution and large modality spectrum variation. In this paper, a novel unsupervised method, named fast projected fuzzy clustering with anchor guidance (FPFC), is proposed for multimodal remote sensing imagery. Specifically, according to the spatial distribution of land covers, meaningful superpixels are obtained for denoising and generating high-quality anchor. The denoised data and anchors are projected into the optimal subspace to jointly learn the shared anchor graph as well as the shared anchor membership matrix from different modalities in an adaptively weighted manner to accelerate the clustering process. Finally, the shared anchor graph and shared anchor membership matrix are combined to derive clustering labels for all pixels. An effective alternating optimization algorithm is designed to solve the proposed formulation. This is the first attempt to propose a soft clustering method for large-scale multimodal remote sensing data. Experiments show that the proposed FPFC achieves 81.34%, 55.43% and 93.34% clustering accuracies on the three datasets and outperforms the state-of-the-art methods. The source code is released at https://github.com/ZhangYongshan/FPFC.

Identifying factors related to pedestrian and cyclist crashes in ACT, Australia with an extended crash dataset.

As vulnerable road users, pedestrians and cyclists are facing a growing number of injuries and fatalities, which has raised increasing safety concerns globally. Based on the crash records collected in the Australian Capital Territory (ACT) in Australia from 2012 to 2021, this research firstly establishes an extended crash dataset by integrating road network features, land use features, and other features. With the extended dataset, we further explore pedestrian and cyclist crashes at macro- and micro-levels. At the macro-level, random parameters negative binomial (RPNB) model is applied to evaluate the effects of Suburbs and Localities Zones (SLZs) based variables on the frequency of pedestrian and cyclist crashes. At the micro-level, binary logit model is adopted to evaluate the effects of event-based variables on the severity of pedestrian and cyclist crashes. The research findings show that multiple factors are associated with high frequency of pedestrian total crashes and fatal/injury crashes, including high population density, high percentage of urban arterial road, low on-road cycleway density, high number of traffic signals and high number of schools. Meanwhile, many factors have positive relations with high frequency of cyclist total crashes and fatal/injury crashes, including high population density, high percentage of residents cycling to work, high median household income, high percentage of households with no motor vehicle, high percentage of urban arterial road and rural road, high number of bus stops and high number of schools. Additionally, it is found that more severe pedestrian crashes occur: (i) at non-signal intersections, (ii) in suburb areas, (iii) in early morning, and (iv) on weekdays. More severe cyclist crashes are observed when the crash type is overturned or struck object/pedestrian/animal; when more than one cyclist is involved; and when crash occurs at park/green space/nature reserve areas.

Intra-articular injection of modified citrus pectin and hyaluronate gel induces synergistic effects in treating osteoarthritis.

We previously found that modified citrus pectin (MCP), an inhibitor of pro-inflammatory factor Galectin-3 (Gal-3), has significant anti-inflammatory and chondroprotective effects. In this study, a hyaluronate (HA) gel-based sustained release system of MCP (MCP-HA) was developed as an anti-inflammatory agent for chronic inflammation for osteoarthritis (OA) treatment. The MCP-HA gel was injected into the knee joint cavities of OA rabbit models induced by anterior cruciate ligament transection (ACLT) or modified Hulth method once a week for five weeks. We found that MCP-HA could improve the symptoms and signs of OA, protect articular cartilage from degeneration, suppress synovial inflammation, and therefore alleviate OA progression. Proteomic analysis of the synovial fluid obtained from the knee joints of OA rabbits revealed that MCP-HA synergistically regulated the levels of multiple inflammatory mediators and proteins involved in metabolic pathways. Taken together, our results demonstrate that the MCP-HA shows a synergistic effect of HA and MCP by modulating both inflammation and metabolic processes, thereby alleviating OA progression. The MCP-HA sustained release system has promising potential for long-term use in OA treatment.

Attention-based Sparse and Collaborative Spectral Abundance Learning for Hyperspectral Subpixel Target Detection.

The subpixel target detection in hyperspectral image processing persists as a formidable challenge. In this paper, we present a novel subpixel target detector termed attention-based sparse and collaborative spectral abundance learning for subpixel target detection in hyperspectral images. To help suppress background during subpixel target detection, the proposed method presents a pixel attention-based background sample selection method for background dictionary construction. Besides, the proposed method integrates a band attention-based spectral abundance learning model, replete with sparse and collaborative constraints, in which the band attention map can contribute to enhancing the discriminative ability of the detector in identifying targets from backgrounds. Ultimately, the detection result of the proposed detector is achieved by the learned target spectral abundance after solving the designed model using the alternating direction method of multipliers algorithm. Rigorous experiments conducted on four benchmark datasets, including one simulated and three real-world datasets, validate the effectiveness of the detector with the probability of detection of 90.88%, 96.86%, and 97.79% on the PHI, RIT Campus, and Reno Urban data, respectively, under fixed false alarm rate equal 0.01, indicating that the proposed method yields superior hyperspectral subpixel detection performance and outperforms existing methodologies.

Federated Learning for Generalization, Robustness, Fairness: A Survey and Benchmark.

Federated learning has emerged as a promising paradigm for privacy-preserving collaboration among different parties. Recently, with the popularity of federated learning, an influx of approaches have delivered towards different realistic challenges. In this survey, we provide a systematic overview of the important and recent developments of research on federated learning. Firstly, we introduce the study history and terminology definition of this area. Then, we comprehensively review three basic lines of research: generalization, robustness, and fairness, by introducing their respective background concepts, task settings, and main challenges. We also offer a detailed overview of representative literature on both methods and datasets. We further benchmark the reviewed methods on several well-known datasets. Finally, we point out several open issues in this field and suggest opportunities for further research. We also provide a public website to continuously track developments in this fast advancing field: https://github.com/WenkeHuang/MarsFL.

An Antibacterial, Conductive Nanocomposite Hydrogel Coupled with Electrical Stimulation for Accelerated Wound Healing.

Background: Electrical stimulation (ES) can effectively promote skin wound healing; however, single-electrode-based ES strategies are difficult to cover the entire wound area, and the effectiveness of ES is often limited by the inconsistent mechanical properties of the electrode and wound tissue. The above factors may lead to ES treatment is not ideal. Methods: A multifunctional conductive hydrogel dressing containing methacrylated gelatin (GelMA), Ti3C2 and collagen binding antimicrobial peptides (V-Os) was developed to improve wound management. Ti3C2 was selected as the electrode component due to its excellent electrical conductivity, the modified antimicrobial peptide V-Os could replace traditional antibiotics to suppress bacterial infections, and GelMA hydrogel was used due to its clinical applicability in wound healing. Results: The results showed that this new hydrogel dressing (GelMA@Ti3C2/V-Os) not only has excellent electrical conductivity and biocompatibility but also has a durable and efficient bactericidal effect. The modified antimicrobial peptides V-Os used were able to bind more closely to GelMA hydrogel to exert long-lasting antibacterial effects. The results of cell experiment showed that the GelMA@Ti3C2/V-Os hydrogel dressing could enhance the effect of current stimulation and significantly improve the migration, proliferation and tissue repair related genes expression of fibroblasts. In vitro experiments results showed that under ES, GelMA@Ti3C2/V-Os hydrogel dressing could promote re-epithelialization, enhance angiogenesis, mediate immune response and prevent wound infection. Conclusion: This multifunctional nanocomposite hydrogel could provide new strategies for promoting infectious wound healing.

Evaluation of TikTok videos on acute pancreatitis: content quality and reliability analysis.

BACKGROUND: Acute pancreatitis (AP) is a common acute digestive system disorder, with patients often turning to TikTok for AP-related information. However, the platform's video quality on AP has not been thoroughly investigated. OBJECTIVE: The main purpose of this study is to evaluate the quality of videos about AP on TikTok, and the secondary purpose is to study the related factors of video quality. METHODS: This study involved retrieving AP-related videos from TikTok, determining, and analyzing them based on predefined inclusion and exclusion criteria. Relevant data were extracted and compiled for evaluation. Video quality was scored using the DISCERN instrument and the Health on the Net (HONcode) score, complemented by introducing the Acute Pancreatitis Content Score (APCS). Pearson correlation analysis was used to assess the correlation between video quality scores and user engagement metrics such as likes, comments, favorites, retweets, and video duration. RESULTS: A total of 111 TikTok videos were included for analysis, and video publishers were composed of physicians (89.18%), news media organizations (13.51%), individual users (5.41%), and medical institutions (0.9%). The majority of videos focused on AP-related educational content (64.87%), followed by physicians' diagnostic and treatment records (15.32%), and personal experiences (19.81%). The mean scores for DISCERN, HONcode, and APCS were 33.05 ± 7.87, 3.09 ± 0.93, and 1.86 ± 1.30, respectively. The highest video scores were those posted by physicians (35.17 ± 7.02 for DISCERN, 3.31 ± 0.56 for HONcode, and 1.94 ± 1.34 for APCS, respectively). According to the APCS, the main contents focused on etiology (n = 55, 49.5%) and clinical presentations (n = 36, 32.4%), followed by treatment (n = 24, 21.6%), severity (n = 20, 18.0%), prevention (n = 19, 17.1%), pathophysiology (n = 17, 15.3%), definitions (n = 13, 11.7%), examinations (n = 10, 9%), and other related content. There was no correlation between the scores of the three evaluation tools and the number of followers, likes, comments, favorites, and retweets of the video. However, DISCERN (r = 0.309) and APCS (r = 0.407) showed a significant positive correlation with video duration, while HONcode showed no correlation with the duration of the video. CONCLUSIONS: The general quality of TikTok videos related to AP is poor; however, the content posted by medical professionals shows relatively higher quality, predominantly focusing on clinical presentations and etiologies. There is a discernible correlation between video duration and quality ratings, indicating that a combined approach incorporating the guideline can comprehensively evaluate AP-related content on TikTok.

Federated Learning With Only Positive Labels by Exploring Label Correlations.

Federated learning (FL) aims to collaboratively learn a model by using the data from multiple users under privacy constraints. In this article, we study the multilabel classification (MLC) problem under the FL setting, where trivial solution and extremely poor performance may be obtained, especially when only positive data with respect to a single class label is provided for each client. This issue can be addressed by adding a specially designed regularizer on the server side. Although effective sometimes, the label correlations are simply ignored and thus suboptimal performance may be obtained. Besides, it is expensive and unsafe to exchange user's private embeddings between server and clients frequently, especially when training model in the contrastive way. To remedy these drawbacks, we propose a novel and generic method termed federated averaging (FedAvg) by exploring label correlations (FedALCs). Specifically, FedALC estimates the label correlations in the class embedding learning for different label pairs and utilizes it to improve the model training. To further improve the safety and also reduce the communication overhead, we propose a variant to learn fixed class embedding for each client, so that the server and clients only need to exchange class embeddings once. Extensive experiments on multiple popular datasets demonstrate that our FedALC can significantly outperform the existing counterparts.

Lipid metabolism disorder in diabetic kidney disease.

Diabetic kidney disease (DKD), a significant complication associated with diabetes mellitus, presents limited treatment options. The progression of DKD is marked by substantial lipid disturbances, including alterations in triglycerides, cholesterol, sphingolipids, phospholipids, lipid droplets, and bile acids (BAs). Altered lipid metabolism serves as a crucial pathogenic mechanism in DKD, potentially intertwined with cellular ferroptosis, lipophagy, lipid metabolism reprogramming, and immune modulation of gut microbiota (thus impacting the liver-kidney axis). The elucidation of these mechanisms opens new potential therapeutic pathways for DKD management. This research explores the link between lipid metabolism disruptions and DKD onset.

Correlation Between Anesthesia Methods and Adverse Short-Term Postoperative Outcomes Depending on Frailty: A Prospective Cohort Study.

Purpose: This study aims to investigate how the type of anesthesia used during major orthopedic surgery may impact adverse short-term postoperative outcomes depending on frailty. Methods: To conduct this investigation, we recruited individuals aged 65 years and older who underwent major orthopedic surgery between March 2022 and April 2023 at a single institution. We utilized the FRAIL scale to evaluate frailty. The primary focus was on occurrences of death or the inability to walk 60 days after the surgery. Secondary measures included death within 60 days; inability to walk without human assistance at 60 days; death or the inability to walk without human assistance at 30 days after surgery, the first time out of bed after surgery, postoperative blood transfusion, length of hospital stay, hospital costs, and the occurrence of surgical complications such as dislocation, periprosthetic fracture, infection, reoperation, wound complications/hematoma. Results: In a study of 387 old adult patients who had undergone major orthopedic surgery, 41.3% were found to be in a frail state. Among these patients, 262 had general anesthesia and 125 had neuraxial anesthesia. Multifactorial logistic regression analyses showed that anesthesia type was not linked to complications. Instead, frailty (OR 4.04, 95% CI 1.04 to 8.57, P< 0.001), age (OR 1.05, 95% CI 1.00-1.10, P= 0.017), and aCCI scores, age-adjusted Charlson Comorbidity Index, (OR 1.36, 95% CI 1.12 to 1.66, P= 0.002) were identified as independent risk factors for death or new walking disorders in these patients 60 days after surgery. After adjusting for frailty, anesthesia methods was not associated with the development of death or new walking disorders in these patients (P > 0.05). Conclusion: In different frail populations, neuraxial anesthesia is likely to be comparable to general anesthesia in terms of the incidence of short-term postoperative adverse outcomes.

Genome-wide association study and genomic prediction for resistance to brown planthopper in rice.

The brown planthopper (BPH) is the most destructive insect pest that threatens rice production globally. Developing rice varieties incorporating BPH-resistant genes has proven to be an effective control measure against BPH. In this study, we assessed the resistance of a core collection consisting of 502 rice germplasms by evaluating resistance scores, weight gain rates and honeydew excretions. A total of 117 rice varieties (23.31%) exhibited resistance to BPH. Genome-wide association studies (GWAS) were performed on both the entire panel of 502 rice varieties and its subspecies, and 6 loci were significantly associated with resistance scores (P value < 1.0e-8). Within these loci, we identified eight candidate genes encoding receptor-like protein kinase (RLK), nucleotide-binding and leucine-rich repeat (NB-LRR), or LRR proteins. Two loci had not been detected in previous study and were entirely novel. Furthermore, we evaluated the predictive ability of genomic selection for resistance to BPH. The results revealed that the highest prediction accuracy for BPH resistance reached 0.633. As expected, the prediction accuracy increased progressively with an increasing number of SNPs, and a total of 6.7K SNPs displayed comparable accuracy to 268K SNPs. Among various statistical models tested, the random forest model exhibited superior predictive accuracy. Moreover, increasing the size of training population improved prediction accuracy; however, there was no significant difference in prediction accuracy between a training population size of 737 and 1179. Additionally, when there existed close genetic relatedness between the training and validation populations, higher prediction accuracies were observed compared to scenarios when they were genetically distant. These findings provide valuable resistance candidate genes and germplasm resources and are crucial for the application of genomic selection for breeding durable BPH-resistant rice varieties.

A Modular Neural Motion Retargeting System Decoupling Skeleton and Shape Perception.

Motion mapping between characters with different structures but corresponding to homeomorphic graphs, meanwhile preserving motion semantics and perceiving shape geometries, poses significant challenges in skinned motion retargeting. We propose M-R 2 ET, a modular neural motion retargeting system to comprehensively address these challenges. The key insight driving M-R 2 ET is its capacity to learn residual motion modifications within a canonical skeleton space. Specifically, a cross-structure alignment module is designed to learn joint correspondences among diverse skeletons, enabling motion copy and forming a reliable initial motion for semantics and geometry perception. Besides, two residual modification modules, i.e., the skeleton-aware module and shape-aware module, preserving source motion semantics and perceiving target character geometries, effectively reduce interpenetration and contact-missing. Driven by our distance-based losses that explicitly model the semantics and geometry, these two modules learn residual motion modifications to the initial motion in a single inference without post-processing. To balance these two motion modifications, we further present a balancing gate to conduct linear interpolation between them. Extensive experiments on the public dataset Mixamo demonstrate that our M-R 2 ET achieves the state-of-the-art performance, enabling cross-structure motion retargeting, and providing a good balance among the preservation of motion semantics, as well as the attenuation of interpenetration and contact-missing.

Clinical efficacy of tumor organoid-guided cancer therapy for locally advanced unresectable or metastatic breast cancer.

Patient-derived organoids (PDOs) may facilitate treatment selection. This retrospective cohort study evaluated the feasibility and clinical benefit of using PDOs to guide personalized treatment in metastatic breast cancer (MBC). Patients diagnosed with MBC were recruited between January 2019 and August 2022. PDOs were established and the efficacy of customized drug panels was determined by measuring cell mortality after drug exposure. Patients receiving organoid-guided treatment (OGT) were matched 1:2 by nearest neighbor propensity scores with patients receiving treatment of physician's choice (TPC). The primary outcome was progression-free survival. Secondary outcomes included objective response rate and disease control rate. Targeted gene sequencing and pathway enrichment analysis were performed. Forty-six PDOs (46 of 51, 90.2%) were generated from 45 MBC patients. PDO drug screening showed an accuracy of 78.4% (95% CI 64.9%-91.9%) in predicting clinical responses. Thirty-six OGT patients were matched to 69 TPC patients. OGT was associated with prolonged median progression-free survival (11.0 months vs. 5.0 months; hazard ratio 0.53 [95% CI 0.33-0.85]; p = .01) and improved disease control (88.9% vs. 63.8%; odd ratio 4.26 [1.44-18.62]) compared with TPC. The objective response rate of both groups was similar. Pathway enrichment analysis in hormone receptor-positive, human epidermal growth factor receptor 2-negative patients demonstrated differentially modulated pathways implicated in DNA repair and transcriptional regulation in those with reduced response to capecitabine/gemcitabine, and pathways associated with cell cycle regulation in those with reduced response to palbociclib. Our study shows that PDO-based functional precision medicine is a feasible and effective strategy for MBC treatment optimization and customization.

Multi-tailed vision transformer for efficient inference.

Recently, Vision Transformer (ViT) has achieved promising performance in image recognition and gradually serves as a powerful backbone in various vision tasks. To satisfy the sequential input of Transformer, the tail of ViT first splits each image into a sequence of visual tokens with a fixed length. Then, the following self-attention layers construct the global relationship between tokens to produce useful representation for the downstream tasks. Empirically, representing the image with more tokens leads to better performance, yet the quadratic computational complexity of self-attention layer to the number of tokens could seriously influence the efficiency of ViT's inference. For computational reduction, a few pruning methods progressively prune uninformative tokens in the Transformer encoder, while leaving the number of tokens before the Transformer untouched. In fact, fewer tokens as the input for the Transformer encoder can directly reduce the following computational cost. In this spirit, we propose a Multi-Tailed Vision Transformer (MT-ViT) in the paper. MT-ViT adopts multiple tails to produce visual sequences of different lengths for the following Transformer encoder. A tail predictor is introduced to decide which tail is the most efficient for the image to produce accurate prediction. Both modules are optimized in an end-to-end fashion, with the Gumbel-Softmax trick. Experiments on ImageNet-1K demonstrate that MT-ViT can achieve a significant reduction on FLOPs with no degradation of the accuracy and outperform compared methods in both accuracy and FLOPs.

Adversarial pair-wise distribution matching for remote sensing image cross-scene classification.

Remarkable achievements have been made in the field of remote sensing cross-scene classification in recent years. However, most methods directly align the entire image features for cross-scene knowledge transfer. They usually ignore the high background complexity and low category consistency of remote sensing images, which can significantly impair the performance of distribution alignment. Besides, shortcomings of the adversarial training paradigm and the inability to guarantee the prediction discriminability and diversity can also hinder cross-scene classification performance. To alleviate the above problems, we propose a novel cross-scene classification framework in a discriminator-free adversarial paradigm, called Adversarial Pair-wise Distribution Matching (APDM), to avoid irrelevant knowledge transfer and enable effective cross-domain modeling. Specifically, we propose the pair-wise cosine discrepancy for both inter-domain and intra-domain prediction measurements to fully leverage the prediction information, which can suppress negative semantic features and implicitly align the cross-scene distributions. Nuclear-norm maximization and minimization are introduced to enhance the target prediction quality and increase the applicability of the source knowledge, respectively. As a general cross-scene framework, APDM can be easily embedded with existing methods to boost the performance. Experimental results and analyses demonstrate that APDM can achieve competitive and effective performance on cross-scene classification tasks.