Network pruning is an important research field aiming at reducing computational costs of neural networks. Conventional approaches follow a fixed paradigm which first trains large and redundant network, then determines units (e.g., channels) are less thus can be removed. In this work, we find that pre-training over-parameterized model not necessary for obtaining the target pruned structure. fact, fully-trained will reduce search space We empirically show more diverse structures directly from...

Face recognition is greatly improved by deep convolutional neural networks (CNNs). Recently, these face models have been used for identity authentication in security sensitive applications. However, CNNs are vulnerable to adversarial patches, which physically realizable and stealthy, raising new concerns on the real-world applications of models. In this paper, we evaluate robustness using patches based transferability, where attacker has limited accessibility target First, extend existing...

As an important auxiliary tool of arrhythmia diagnosis, Electrocardiogram (ECG) is frequently utilized to detect a variety cardiovascular diseases caused by arrhythmia, such as cardiac mechanical infarction. In the past few years, classification ECG has always been challenging problem. This paper presents novel deep learning model called convolutional vision transformer (ConViT), which combines (ViT) with neural network (CNN), for classification, in unique soft inductive bias gated...

One critical challenge in 6D object pose estimation from a single RGBD image is efficient integration of two different modalities, i.e., color and depth. In this work, we tackle problem by novel Deep Fusion Transformer (DFTr) block that can aggregate cross-modality features for improving estimation. Unlike existing fusion methods, the proposed DFTr better model semantic correlation leveraging their similarity, such globally enhanced modalities be integrated improved information extraction....

Point-of-Interest (POI) recommendation has been extensively studied and successfully applied in industry recently. However, most existing approaches build centralized models on the basis of collecting users’ data. Both private data are held by recommender, which causes serious privacy concerns. In this article, we propose a novel Privacy preserving POI Recommendation (PriRec) framework. First, to protect privacy, (features actions) kept their own side, e.g., Cellphone or Pad. Meanwhile,...

Although object detection techniques have been widely employed in various practical applications, automatic tree is still a difficult challenge, especially for street-view images. In this article, we propose unified end-to-end trainable network street based on state-of-the-art deep learning-based detector. We tackle low illumination and heavy occlusion conditions detection, which not extensively studied until now, due to clear challenges. Existing generic detectors cannot be directly applied...

Incorporating review information into the recommender system has been demonstrated to be an effective method for boosting recommendation performance. Previous research mainly focus on designing advanced architectures better profile users and items. However, in realities can highly sparse imbalanced, which poses great challenges user/item representations satisfied performance enhancement. To alleviate this problem, paper, we propose improve review-based by counterfactually augmenting training...

The 6-D pose estimation is a crucial task in vision-based measurement for robotic manipulation. It becomes challenging because of the variety lighting conditions, cluttered background, occlusion, and texture-less objects. various conditions objects lead to dramatic changes imaging. In this article, we propose an edge-attention network (EANet) that achieve autonomous perception edge, which invariant when change are texture-less. To this, EANet adopts multitask learning strategy introduces...

Privacy-preserving machine learning has drawn increasingly attention recently, especially with kinds of privacy regulations come into force. Under such situation, Federated Learning (FL) appears to facilitate privacy-preserving joint modeling among multiple parties. Although many federated algorithms have been extensively studied, there is still a lack secure and practical gradient tree boosting models (e.g., XGB) in literature. In this paper, we aim build large-scale XGB under vertically...

Few-shot classification (FSC) is a fundamental yet challenging task in computer vision that involves recognizing novel classes from limited data. While previous methods have focused on enhancing visual features or incorporating additional modalities, Large Vision Language Models (LVLMs) offer promising alternative due to their rich knowledge and strong perception. However, LVLMs risk learning specific response formats rather than effectively extracting useful information support data FSC. In...

Liver registration by overlaying preoperative 3D models onto intraoperative 2D frames can assist surgeons in perceiving the spatial anatomy of liver clearly for a higher surgical success rate. Existing methods rely heavily on anatomical landmark-based workflows, which encounter two major limitations: 1) ambiguous landmark definitions fail to provide efficient markers registration; 2) insufficient integration visual information shape deformation modeling. To address these challenges, this...

Unmanned Aerial Vehicle (UAV) spray has been used for efficient and adaptive pesticide applications with its low costs. However, droplet drift is the main problem UAV will induce waste safety concerns. Droplet size deposition distribution are both highly related to effect, which determined by nozzle. Therefore, it necessary propose an evaluating method a specific nozzles. In this paper, four machine learning methods (REGRESS, least squares support vector machines (LS-SVM), extreme machine,...

Mainly for the sake of solving lack keyword-specific data, we propose one Keyword Spotting (KWS) system using Deep Neural Network (DNN) and Connectionist Temporal Classifier (CTC) on power-constrained small-footprint mobile devices, taking full advantage general corpus from continuous speech recognition which is great amount. DNN to directly predict posterior phoneme units any personally customized key-phrase, CTC produce a confidence score given sequence as responsive decision-making...

Analytical redundancy technique is of great importance to guarantee the reliability and safety aircraft engine system. In this paper, a machine learning based aeroengine sensor analytical developed verified through hardware-in-the-loop (HIL) simulation. The modified online sequential extreme machine, selective updating regularized (SROS-ELM), employed train model estimate measurements. It selectively updates output weights neural networks according prediction accuracy norm weight vector,...

Classification of data with imbalanced class distribution has encountered a significant drawback by most conventional classification learning methods which assume relatively balanced distribution. This paper proposes novel method based on data-partition and SMOTE for learning. The proposed differs from ones in both the prediction stages. For stage, uses following three steps to learn class-imbalance oriented model: (1) partitioning majority into several clusters using partition such as...

Tool Condition Monitoring (TCM) is an important topic in manufacturing industry, which improves product quality, production efficiency, reduces costs and downtime. This paper develops a new data-driven framework for estimating tool remaining useful life (RUL) TCM. The includes the following modular components: data preprocessing with proposed adaptive Baysian change point detection (ABCPD) automatic alignment, time window process, feature extraction, selection multi-layer neural network as...

Segmentation and classification are important tasks in remote sensing image analysis. Recent research shows that images can be described hierarchical structure or regions. Such hierarchies produce the state-of-the-art segmentations used classification. However, they often contain more levels regions than required for an efficient description, which may cause increased computational complexity. In this letter, we propose a new segmentation method applies graph Laplacian energy as generic...

Bayesian deep learning is recently regarded as an intrinsic way to characterize the weight uncertainty of neural networks (DNNs). Stochastic Gradient Langevin Dynamics (SGLD) effective method enable on large-scale datasets. Previous theoretical studies have shown various appealing properties SGLD, ranging from convergence generalization bounds. In this paper, we study SGLD a novel perspective membership privacy protection (i.e., preventing attack). The attack, which aims determine whether...

The rapid determination of nitrogen, phosphorus, potassium and other major nutrient elements is an important technical guarantee in the quality control chemical fertilizers. In this study, a small visible spectrometer near-infrared were used to collect spectrum information 33 different common fertilizers including compound fertilizers, blended controlled-release 550~950 nm 1050~1640 spectra with stable signals intercepted as analysis spectrum, competitive adaptive reweighted sampling...