Long-tailed data is still a big challenge for deep neural networks, even though they have achieved great success on balanced data. We observe that vanilla training longtailed with crossentropy loss makes the instance-rich head classes severely squeeze spatial distribution of tail classes, which leads to difficulty in classifying class samples. Furthermore, original can only propagate gradient short-lively because softmax form rapidly approaches zero as logit difference increases. This...

Deep neural networks have made huge progress in the last few decades. However, as real-world data often exhibits a long-tailed distribution, vanilla deep models tend to be heavily biased toward majority classes. To address this problem, state-of-the-art methods usually adopt mixture of experts (MoE) focus on different parts distribution. Experts these are with same model depth, which neglects fact that classes may preferences fit by depths. end, we propose novel MoE-based method called...

The imbalanced distribution of long-tailed data presents a considerable challenge for deep learning models, as it causes them to prioritize the accurate classification head classes but largely disregard tail classes. biased decision boundary caused by inadequate semantic information in is one key factors contributing their low recognition accuracy. To rectify this issue, we propose augment grafting diverse from classes, referred head-to-tail fusion (H2T). We replace portion feature maps with...

The selection of training samples is important for the accuracy and efficiency synthetic aperture radar (SAR) image change detection task. However, are traditionally extracted from whole image, which leads to longer time an unbalanced number pixels in changed unchanged classes. To overcome this problem, we propose a novel method combining saliency with principal component analysis network, named SDPCANet. enhance reliability reduce amount samples, SDPCANet uses context-aware obtain salient...

Cross-modal hashing (CMH) has attracted considerable attention in recent years. Almost all existing CMH methods primarily focus on reducing the modality gap and semantic gap, i.e., aligning multi-modal features their semantics Hamming space, without taking into account space difference between real number space. In fact, can affect performance of methods. this paper, we analyze demonstrate how affects methods, which therefore raises two problems: solution compression loss function...

Numerous field tests indicate that the soil–structure interaction (SSI) has a significant impact on dynamic characteristics of super-tall buildings, which may lead to unexpected structural seismic responses and/or failure. Taking Shanghai Tower with total height 632 m as research object, substructure approach is used simulate SSI effect Tower. The refined finite element (FE) model superstructure and simplified analytical foundation adjacent soil are established. Subsequently, collapse...

This paper presents multisensor image fusion of polarization and infrared imaging to detect defects in printed circuit boards (PCBs). Many existing automated optical inspection techniques rely on visible sensors. However, collected images suffer from uneven brightness levels due the influence lighting environment, which may significantly affect detection accuracies. Polarization information characterizes material types, surface roughness, geometric shape an object. Thermal reveals heat...

It is not uncommon that real-world data are distributed with a long tail. For such data, the learning of deep neural networks becomes challenging because it hard to classify tail classes correctly. In literature, several existing methods have addressed this problem by reducing classifier bias, provided features obtained long-tailed representative enough. However, we find training directly on leads uneven embedding space. That is, space head severely compresses classes, which conducive...

Federated Clustering (FC) is crucial to mining knowledge from unlabeled non-Independent Identically Distributed (non-IID) data provided by multiple clients while preserving their privacy. Most existing attempts learn cluster distributions at local clients, then securely pass the desensitized information server for aggregation. However, some tricky but common FC problems are still relatively unexplored, including heterogeneity in terms of clients' communication capacity and unknown number...

With ongoing development of earthquake engineering research and the lessons learnt from a series strong earthquakes, seismic design concept "resilience" has received much attention. Resilience describes capability structure or city to recover rapidly after earthquakes other disasters. As one main features urban constructions, tall buildings have greater impact on sustainability resilience major cities. Therefore, it is important timely quantify their resilience. In this work, quantitative...

For long-tailed distributed data, existing classification models often learn overwhelmingly on the head classes while ignoring tail classes, resulting in poor generalization capability. To address this problem, we thereby propose a new approach paper, which key point sensitive (KPS) loss is presented to regularize points strongly improve performance of model. Meanwhile, order proposed KPS also assigns relatively large margins classes. Furthermore, gradient adjustment (GA) optimization...

To address the current demands for antenna miniaturization, ultra-bandwidth, and circular polarization in advanced medical devices, a novel ISM band implantable blood glucose monitoring has been developed. This achieves miniaturization by incorporating slots radiation patch adding symmetric short-circuit probes, resulting compact size of only 0.054λ0 × 0.005λ0 (λ0 is wavelength free space respect lowest working frequency). By combining two resonance points utilizing differential feed...

Abstract Accurate evaluation of lattice thermal conductivity is usually a tough task from the theoretical side, especially for alloyed systems with fractional stoichiometry. Using tetradymite family as prototypical class examples, we propose reliable approach rapid prediction on at arbitrary composition by utilizing concept configurational entropy. Instead performing time-consuming first-principles calculations, conductivities any tetradymites can be readily obtained few samples integer The...

Supervised cross-modal hashing has received wide attention in recent years. However, existing methods primarily rely on sample-wise semantic relationships to evaluate the similarity between samples, overlooking impact of label distribution enhancing retrieval performance. Moreover, limited representation capability traditional dense hash codes hinders preservation relationship. To overcome these challenges, we propose a new method, Joint Semantic Preserving Sparse Hashing (JSPSH)....

This article focuses on a new and practical problem in single-sample per person face recognition (SSPP FR), i.e., SSPP FR with contaminated biometric enrolment database (SSPP-ce where the SSPP-based is by nuisance facial variations wild, such as poor lightings, expression change, disguises (e.g., wearing sunglasses, hat, scarf). In SSPP-ce FR, most popular generic learning methods will suffer serious performance degradation because prototype plus variation (P+V) model used these no longer...

Deep neural networks frequently suffer from performance degradation when the training data is long-tailed because several majority classes dominate training, resulting in a biased model. Recent studies have made great effort solving this issue by obtaining good representations space, but few of them pay attention to influence feature norm on predicted results. In paper, we therefore address problem space and thereby propose feature-balanced loss. Specifically, encourage larger norms tail...

As an effective tool for network compression, pruning techniques have been widely used to reduce the large number of parameters in deep neural networks (NNs). Nevertheless, unstructured has limitation dealing with sparse and irregular weights. By contrast, structured can help eliminate this drawback but it requires complex criteria determine which components be pruned. Therefore, paper presents a new method termed BUnit-Net, directly constructs compact NNs by stacking designed basic units,...