RESUMEN

Detecting Out-of-Distribution (OOD) inputs is essential for reliable deep learning in the open world. However, most existing OOD detection methods have been developed based on training sets that exhibit balanced class distributions, making them susceptible when confronted with training sets following a long-tailed distribution. To alleviate this problem, we propose an effective three-branch training framework, which demonstrates the efficacy of incorporating an extra rejection class along with auxiliary outlier training data for effective OOD detection in long-tailed image classification. In our proposed framework, all outlier training samples are assigned the label of the rejection class. We employ an inlier loss, an outlier loss, and a Tail-class prototype induced Supervised Contrastive Loss (TSCL) to train both the in-distribution classifier and OOD detector within one network. During inference, the OOD detector is constructed using the rejection class. Extensive experimental results demonstrate that the superior OOD detection performance of our proposed method in long-tailed image classification. For example, in the more challenging case where CIFAR100-LT is used as in-distribution, our method improves the average AUROC by 1.23% and reduces the average FPR95 by 3.18% compared to the baseline method utilizing Outlier Exposure (OE). Code is available at github.

Asunto(s)

RESUMEN

As the deployment of artificial intelligence (AI) models in real-world settings grows, their open-environment robustness becomes increasingly critical. This study aims to dissect the robustness of deep learning models, particularly comparing transformer-based models against CNN-based models. We focus on unraveling the sources of robustness from two key perspectives: structural and process robustness. Our findings suggest that transformer-based models generally outperform convolution-based models in robustness across multiple metrics. However, we contend that these metrics may not wholly represent true model robustness, such as the mean of corruption error. To better understand the underpinnings of this robustness advantage, we analyze models through the lens of Fourier transform and game interaction. From our insights, we propose a calibrated evaluation metric for robustness against real-world data, and a blur-based method to enhance robustness performance. Our approach achieves state-of-the-art results, with mCE scores of 2.1% on CIFAR-10-C, 12.4% on CIFAR-100-C, and 24.9% on TinyImageNet-C.

Asunto(s)

RESUMEN

Outliers will inevitably creep into the captured point cloud during 3D scanning, degrading cutting-edge models on various geometric tasks heavily. This paper looks at an intriguing question that whether point cloud completion and segmentation can promote each other to defeat outliers. To answer it, we propose a collaborative completion and segmentation network, termed CS-Net, for partial point clouds with outliers. Unlike most of existing methods, CS-Net does not need any clean (or say outlier-free) point cloud as input or any outlier removal operation. CS-Net is a new learning paradigm that makes completion and segmentation networks work collaboratively. With a cascaded architecture, our method refines the prediction progressively. Specifically, after the segmentation network, a cleaner point cloud is fed into the completion network. We design a novel completion network which harnesses the labels obtained by segmentation together with farthest point sampling to purify the point cloud and leverages KNN-grouping for better generation. Benefited from segmentation, the completion module can utilize the filtered point cloud which is cleaner for completion. Meanwhile, the segmentation module is able to distinguish outliers from target objects more accurately with the help of the clean and complete shape inferred by completion. Besides the designed collaborative mechanism of CS-Net, we establish a benchmark dataset of partial point clouds with outliers. Extensive experiments show clear improvements of our CS-Net over its competitors, in terms of outlier robustness and completion accuracy.

RESUMEN

Most multimodal learning methods assume that all modalities are always available in data. However, in real-world applications, the assumption is often violated due to privacy protection, sensor failure etc. Previous works for incomplete multimodal learning often suffer from one of the following drawbacks: introducing noise, lacking flexibility to missing patterns and failing to capture interactions between modalities. To overcome these challenges, we propose a COntrastive Masked-attention model (COM). The framework performs cross-modal contrastive learning with GAN-based augmentation to reduce modality gap, and employs a masked-attention model to capture interactions between modalities. The augmentation adapts cross-modal contrastive learning to suit incomplete case by a two-player game, improving the effectiveness of multimodal representations. Interactions between modalities are modeled by stacking self-attention blocks, and attention masks limit them on the observed modalities to avoid extra noise. All kinds of modality combinations share a unified architecture, so the model is flexible to different missing patterns. Extensive experiments on six datasets demonstrate the effectiveness and robustness of the proposed method for incomplete multimodal learning.

Asunto(s)

RESUMEN

Graph Neural Networks (GNNs) have achieved state-of-the-art performance in graph-related tasks. For graph classification task, an elaborated pooling operator is vital for learning graph-level representations. Most pooling operators derived from existing GNNs generate a coarsen graph through ordering the nodes and selecting some top-ranked ones. However, these methods fail to explore the fundamental elements other than nodes in graphs, which may not efficiently utilize the structure information. Besides, all edges attached to the low-ranked nodes are discarded, which destroys graphs' connectivity and loses information. Moreover, the selected nodes tend to concentrate on some substructures while overlooking information in others. To address these challenges, we propose a novel pooling operator called Accurate Structure-Aware Graph Pooling (ASPool), which can be integrated into various GNNs to learn graph-level representation. Specifically, ASPool adaptively retains a subset of edges to calibrate the graph structure and learns the abstracted representations, wherein all the edges are viewed as non-peers instead of simply connecting nodes. To preserve the graph's connectivity, we further introduce the selection strategy considering both top-ranked nodes and dropped edges. Additionally, ASPool performs a two-stage calculation process to promise that the sampled nodes are distributed throughout the graph. Experiment results on 9 widely used benchmarks show that ASPool achieves superior performance over the state-of-the-art graph representation learning methods.

Asunto(s)

RESUMEN

Objective: To analyze treatment strategies, prognosis, and related risk factors of patients with postinfarction ventricular septal rupture, as well as the impact of timing of surgical intervention. Methods: A total of 23 patients diagnosed with postinfarction ventricular septal rupture who were non-selectively admitted to Shanxi Provincial Cardiovascular Hospital between October 2017 and August 2021 were included in this study. The relevant clinical data, operation-related conditions, and follow-up data were summarized for all patients. The Kaplan-Meier method and log-rank test were used for the cumulative incidence of unadjusted mortality in patients with different treatment methods. Multivariate logistic regression was used to evaluate the independent risk factors for in-hospital patient mortality. Results: The mean age of the study patients was 64.43 ± 7.54 years, 12(52.2%) were females. There was a significant difference in terms of postoperative residual shunt between the surgical and interventional closure groups (5.9 vs. 100%, respectively; P < 0.001). The overall in-hospital mortality rate was 21.7%; however, even though the surgical group had a lower mortality rate than the interventional closure group (17.6 vs. 33%, respectively), this difference was not statistically significant (P = 0.576). Univariate analysis showed that in-hospital survival group patients were significantly younger than in-hospital death group patients (62.50 ± 6.53 vs. 71.40 ± 7.37 years, respectively; P = 0.016), and that women had a significantly higher in-hospital mortality rate than men (P = 0.037). The average postoperative follow-up time was 18.11 ± 13.92 months; as of the end of the study all 14 patients in the surgical group were alive, Two out of four patients survived and two patients died after interventional closure. Univariate analysis showed that interventional closure was a risk factor for long-term death (P < 0.05). Conclusion: Surgical operation is the most effective treatment for patients with postinfarction ventricular septal rupture; however, the best timing of the operation should be based on the patient's condition and comprehensively determined through real-time evaluation and monitoring. We believe that delaying the operation time as much as possible when the patient's condition permits can reduce postoperative mortality. Interventional closure can be used as a supplementary or bridge treatment for surgical procedures.

RESUMEN

BACKGROUND: Several models have been developed to predict asymptomatic carotid stenosis (ACS), however these models did not pay much attention to people with lower level of stenosis (<50% or carotid plaques, especially instable carotid plaques) who might benefit from early interventions. Here, we developed a new model to predict unstable carotid plaques through systematic screening in population with high risk of stroke. METHODS: Community residents who participated the China National Stroke Screening and Prevention Project (CNSSPP) were screened for their stroke risks. A total of 2841 individuals with high risk of stroke were enrolled in this study, 266 (9.4%) of them were found unstable carotid plaques. A total of 19 risk factors were included in this study. Subjects were randomly distributed into Derivation Set group or Validation Set group. According to their carotid ultrasonography records, subjects in derivation set group were further categorized into unstable plaque group or stable plaque group. RESULTS: 174 cases and 1720 cases from Derivation Set group were categorized into unstable plaque group and stable plaque group respectively. The independent risk factors for carotid unstable plaque were: male (OR 1.966, 95%CI 1.406-2.749), older age (50-59, OR 6.012, 95%CI 1.410-25.629; 60-69, OR 13.915, 95%CI 3.381-57.267;≥70, OR 31.267, 95%CI 7.472-130.83), married(OR 1.780, 95%CI 1.186-2.672), LDL-C(OR 2.015, 95%CI 1.443-2.814), and HDL-C(OR 2.130, 95%CI 1.360-3.338). A predictive scoring system was generated, ranging from 0 to 10. The cut-off value of this predictive scoring system is 6.5. The AUC value for derivation and validation set group were 0.738 and 0.737 respectively. CONCLUSIONS: For those individuals with high risk of stroke, we developed a new model which could identify those who have a higher chance to have unstable carotid plaques. When an individual's predictive model score exceeds 6.5, the probability of having carotid unstable plaques is high, and carotid ultrasonography should be conducted accordingly. This model could be helpful in the primary prevention of stroke.

Asunto(s)

RESUMEN

A uranyl detection strategy with ultra-sensitivity was developed based on entropy-driven amplification and DNAzyme circular cleavage amplification. The cleavage of UO22+-specific DNAzyme produces a DNA fragment to initiate the entropy-driven amplification. Two DNA sequences released from the entropy-driven amplification are partly complementary. They can form an entire enzyme strand (E-DNA) of Mg2+-specific DNAzyme. The formed E-DNA can circularly cleave FAM-labeled probes on gold nanoparticles (AuNPs), causing the leaving of FAM from AuNPs and recovery of fluorescent signal. A linear relationship was obtained in the range from 30 pM to 5 nM between fluorescence intensity and concentration of UO22+. The limit of detection was low to 13 pM. This method showed a promising future for practical application in real water samples.

Asunto(s)

RESUMEN

E-cigarettes (vape) are now the most commonly used tobacco product among youth in the United States. Ads are claiming e-cigarettes help smokers quit, but most of them contain nicotine, which can cause addiction and harm the developing adolescent brain. Therefore, national, state, and local health organizations have proposed anti-vaping campaigns to warn the potential risks of e-cigarettes. However, there is some evidence that these products may reduce harm for adult users who reduce or quit combustible cigarette smoking, and with little evidence that e-cigarettes cause long-term harm, pro-vaping advocates have used this equivocal evidence base to oppose the anti-vaping media campaign messaging, generating a very high volume of oppositional messages on social media. Thus, when we analyze the feedback of anti-vaping campaigns, it is crucial to partition the audience into different clusters according to their attitudes and affiliations. Motivated by this, in this paper, we propose the "community detection on anti-vaping campaign audience" problem and design the "community detection based on social, repost and content relation, (Sorento)" algorithm to solve it. Sorento computes users' intimacy scores based on their social connections, repost relations, and content similarities. The community detection results achieved by Sorento demonstrate that though anti-vaping campaigns are proposed in different areas at different times, their opponent messages are mainly posted by the same community of pro-vapors.

Asunto(s)

RESUMEN

In order to explore the compatible principles of Xiebai decoction family, formulae from ancient and modern Xiebai decoction family were collected and sorted in this study. The compatible characteristics, core herbs, as well as the relativity of herbs nature in Xiebai decoction family were analyzed based on scale free network and other data-mining methods such as association rules, clustering analysis and correspondence analysis. The scale free network results showed that in Xiebai decoction family, Mori Cortex-Lycii Cortex-Glycyrrhizae Radix et Rhizoma was used as the core compatible group and formed the complicated compatible network with other additional herbs; association rules results showed that the core herbs in such formulae included Mori Cortex, Lycii Cortex, Glycyrrhizae Radix et Rhizoma, scutellaria root, Platycodon root, Anemarrhena, and almond, which formed corresponding herbal pairs and compatibility; clustering analysis showed that Mori Cortex was the core herb in Xiebai decoction family, and Mori Cortex-Lycii Cortex-Glycyrrhizae Radix et Rhizoma was its main combination unit, which was always compatible with herbs of clearing heat, reducing phlegm, supplementing Qi and nourishing Yin to form the series prescriptions. The results indicated that the core compatibility features of Xiebai decoction family were clearing heat in lung and relieving cough and asthma, providing a basis for the clinical application of Xiebai decoction family.

Asunto(s)

RESUMEN

OBJECTIVE: To investigate the laws of eighteen incompatible medicaments of the chest pain prescriptions based on association rules mining. METHOD: The database of chest pain prescription was established and then the chest pain prescriptions composed of eighteen incompatible medicaments were screened. The dynasty, couplet medicines, the property and flavor of drugs and preparation form were analyzed with the frequent item sets and corresponding analysis methods. RESULT: Eight hundred and fifty chest pain prescriptions were collected, and 88 of them contained eighteen incompatible medicaments, taking 10.3% of all; the applications of ancient and modern chest pain prescriptions containing eighteen incompatible medicaments are significant difference (P < 0.05). Ancient formulas, mainly focus on the eastern jin dynasty and tang dynasty, are more often used than the modern formulas. The most common anti-drugs is on the Fuzi-Pinellia, Chuanwu-Pinellia; the property and flavor of drugs is bitter cold most closely; the decoction of the formulas is mostly used. CONCLUSION: Eighteen incompatible medicaments account for about ten percent of the chest pain prescription, not an uncommon side. There are certain rules for application of anti-drug compatibility to treat chest pain.

Asunto(s)