We show that adversarial examples, i.e., the visually imperceptible perturbations result in Convolutional Neural Networks (CNNs) fail, can be alleviated with a mechanism based on foveations---applying CNN different image regions. To see this, first, we report results ImageNet lead to revision of hypothesis are consequence CNNs acting as linear classifier: act locally linearly changes regions objects recognized by CNN, and other may non-linearly. Then, corroborate when neural responses...

Thyroid nodules are one of the most common nodular lesions. The incidence thyroid cancer has increased rapidly in past three decades and is cancers with highest incidence. As a non-invasive imaging modality, ultrasonography can identify benign malignant nodules, it be used for large-scale screening. In this study, inspired by domain knowledge sonographers when diagnosing ultrasound images, local global feature disentangled network (LoGo-Net) proposed to classify nodules. This model imitates...

Fairness (also known as equity interchangeably) in machine learning is important for societal well-being, but limited public datasets hinder its progress. Currently, no dedicated medical with imaging data fairness are available, though underrepresented groups suffer from more health issues. To address this gap, we introduce Harvard Glaucoma (Harvard-GF), a retinal nerve disease dataset including 3,300 subjects both 2D and 3D balanced racial glaucoma detection. the leading cause of...

Current research on the association between demographic variables and dietary patterns with atherosclerotic cardiovascular disease (ASCVD) is limited in breadth depth. This study aimed to construct a machine learning (ML) algorithm that can accurately transparently establish correlations variables, habits, ASCVD. The dataset used this originates from United States National Health Nutrition Examination Survey (U.S. NHANES) spanning 1999–2018. Five ML models were developed predict ASCVD,...

Cervical cancer is one of the illness which threatening women's health all over world and it hard to observe any sign in early stage. Three methods have been introduced this paper analyze dataset cervical cancer, including SVM (Support Vector Machine), XGBoost (eXtreme Gradient Boosting) Random Forest. The contains 32 risk factors four target variables: Hinselmann, Schiller, Cytology, Biopsy. And diagnostic results these variables were classified by three that mentioned above. Finally, top...

Intraclass compactness and interclass separability are crucial indicators to measure the effectiveness of a model produce discriminative features, where intraclass indicates how close features with same label each other far away different labels are. In this paper, we investigate learned by convolutional networks propose Gaussian-based softmax ( G -softmax) function that can effectively improve separability. The proposed is simple implement easily replace function. We evaluate -softmax on...

One of the well-known challenges in computer vision tasks is visual diversity images, which could result an agreement or disagreement between learned knowledge and content exhibited by current observation. In this work, we first define such a concepts learning process as congruency. Formally, given particular task sufficiently large dataset, congruency issue occurs whereby task-specific semantics training data are highly varying. We propose Direction Concentration Learning (DCL) method to...

Prognostication is an essential tool for risk adjustment and decision making in the intensive care units (ICUs). In order to improve patient outcomes, we have been trying develop a more effective model than Acute Physiology Chronic Health Evaluation (APACHE) II measure severity of patients ICUs. The aim present study was provide mortality prediction ICUs patients, assess its performance relative based on APACHE scoring system.We used Medical Information Mart Intensive Care version III...

One of the most significant challenges in multi-label image classification is learning representative features that capture rich semantic information a cluttered scene. As an bottleneck, visual attention mechanism allows humans to selectively process important input, enabling rapid and accurate scene understanding. In this work, we study correlation between classification, exploit extra pathway for improving performance. Specifically, propose dual-stream neural network consists two...

A significant body of literature on saliency modeling predicts where humans look in a single image or video. Besides the scientific goal understanding how information is fused from multiple visual sources to identify regions interest holistic manner, there are tremendous engineering applications multi-camera due widespread cameras. This paper proposes principled framework smoothly integrate views global scene map, and employ algorithm incorporating high-level features most important by...

With the continuous improvement of people’s requirements for interactive experience, gesture recognition is widely used as a basic human-computer interaction. However, due to environment, light source, cover, and other factors, diversity complexity gestures have great impact on recognition. In order enhance features recognition, firstly, hand skin color filtered through YCbCr space separate region be recognized, Gaussian filter process noise edge; secondly, morphological gray open operation...

Formal model and verification techniques can be used to design analyze many kinds of systems. However, there are few formal models for analysis sensor networks in the literature. In this paper, we present a graphical simulation system modeling networks. A is proposed based on Space Time Petri Net (STPN). STPN language which extended from time Nets (TPN) colored nets (CPN). The idea add space information places original TPN networks, such as locations nodes. And idea, set new concepts...

Thorax disease classification is a challenging task due to complex pathologies and subtle texture changes, etc. It has been extensively studied for years largely because of its wide application in computer-aided diagnosis. Most existing methods directly learn global feature representations from whole Chest X-ray (CXR) images, without considering depth the richer visual cues lying around informative local regions. Thus, these often produce sub-optimal thorax performance they ignore very...

Recently, an increasing number of works have proposed to learn visual saliency by leveraging human fixations. However, the collection fixations is time consuming and existing eye tracking datasets are generally small when compared with other domains. Thus, it contains a certain degree dataset bias due large image variations (e.g., outdoor scenes vs. emotion-evoking images). In learning based prediction literature, most models trained evaluated within same cross validation not yet common...

Abstract Objective To develop an equitable artificial intelligence model for glaucoma screening. Design Cross-sectional study. Participants 7,418 optical coherence tomography (OCT) paired with reliable visual field (VF) measurements of patients from the Massachusetts Eye and Ear Glaucoma Service between 2021 2023. Methods We developed fair identify normalization (FIN) module to equalize feature importance across different identity groups improve performance equity. EfficientNet served as...

Facial expression recognition (FER) has applications in many scenarios, making it a valuable research direction. However, due to the uncertainty of real-world images, their accuracy is not satisfying. To deal with this problem, we propose two attention modules-based network that uses different modules extract features. The main classifier self-attention module (SAM), and auxiliary channel enhancement module. At same time, classification results classifiers are constrained by variance...

<p>Currently researchers in the field of personalized recommendations bear little consideration on users' interest differences resource attributes although attribute is usually one most important factors determining user preferences. To solve this problem, paper builds an evaluation model based multi-attributes, proposes a modified Pearson-Compatibility multi-attribute group decision-making algorithm, and introduces algorithm to recommendation problem k-neighbor similar users....