Anomaly detection in videos aims at reporting anything that does not conform the normal behaviour or distribution. However, due to sparsity of abnormal video clips real life, collecting annotated data for supervised learning is exceptionally cumbersome. Inspired by practicability generative models semi-supervised learning, we propose a novel sequential model based on variational autoencoder (VAE) future frame prediction with convolutional LSTM (ConvLSTM). To best our knowledge, this first...

Predictive maintenance (PdM) is useful for engineers to schedule flexibly, operate equipment efficiently, and also avoid unexpected downtime. Remaining life (RUL) prediction critical PdM before the need component replacement. Data-driven approaches have attracted more attention RUL in flexible production. Compared with statistical-based conventional machine learning approaches, deep learning-based can extract features from raw sensor data without prior knowledge a domain expert. This study...

With the development of smart manufacturing, in order to detect abnormal conditions equipment, a large number sensors have been used record variables associated with production equipment. This study focuses on prediction Remaining Useful Life (RUL). RUL is part predictive maintenance, which uses trend machine predict when will malfunction. High accuracy not only reduces consumption manpower and materials, but also need for future maintenance. detecting faults as early possible, before needs...

The management of gas pipelines requires efficient, predictive systems to enhance safety and operational efficiency. limitations include potential data quality issues, sensor accuracy, environmental factors that may affect model performance. To develop an intelligent pipeline system by integrating the Wombat Algorithm-driven Scalable Random Forest (WA-SRF) for improvement scalability, fault detection in performance operations pipelines, various sensors are embedded pipeline, collected from...

Data similarity is a key concept in many data-driven applications. Many algorithms are sensitive to measures. To tackle this fundamental problem, automatically learning of information from data via self-expression has been developed and successfully applied various models, such as low-rank representation, sparse subspace learning, semisupervised learning. However, it just tries reconstruct the original some valuable information, e.g., manifold structure, largely ignored. In paper, we argue...

We address the problem of image-based crowd counting. In particular, we propose a new called <i>unlabeled scene-adaptive counting</i>. Given target scene, would like to have counting model specifically adapted this particular scene based on data that capture some information about scene. paper, use one or more unlabeled images from perform adaptation. comparison with existing setups (e.g. fully supervised), our proposed setup is closer real-world applications systems. introduce novel...

Automated Optical Inspection (AOI) is used for defect inspection during industrial manufacturing process. It uses optical instrument to snap the surface of products and identify defects through technique machine vision processing. Deep learning convolution neural network automatically produce feature which are useful correctly. However, class imbalance number samples normal typically in process, will lead poor accuracy deep model. This paper integrate pix2pix, a Conditional Generative...

Entity linking, a crucial task in the realm of natural language processing, aims to link entity mentions text their corresponding entities knowledge base. While long documents provide abundant contextual information, facilitating feature extraction for identification and disambiguation, linking Chinese short texts presents significant challenges. This study introduces an innovative approach within texts, combining multiple embedding representations. It integrates representations from both...

Military information extraction is an important means to gain military advantage. named entity recognition the basis of extraction. For entities, a method supervised based on deep learning was proposed identify and extract entities in texts such as troops, geographical locations, weapons so on. This avoids complexity artificial construction features inaccuracy text segmentation. It uses BI-LSTM CRF model automatic character embedding, then identifies entities. Experiments show that this...

In the field of human–computer interaction, millimeter-wave radar has attracted considerable attention as a contactless and private approach to hand gesture recognition. However, single-target recognition scenes are generally simplistic not representative real interactions. Therefore, this research examines feasibility using single recognize gestures in double-target by combining theory with deep learning. First, dynamic range angle image (DRAI) is composed weights DRAIs two gestures. Thus,...

Data similarity is a key concept in many data-driven applications. Many algorithms are sensitive to measures. To tackle this fundamental problem, automatically learning of information from data via self-expression has been developed and successfully applied various models, such as low-rank representation, sparse subspace learning, semi-supervised learning. However, it just tries reconstruct the original some valuable information, e.g., manifold structure, largely ignored. In paper, we argue...

Abstract Military named entity recognition is the basis of military intelligence analysis and operational information service. In order to solve problems inaccurate word segmentation, diverse forms lack corpus in texts, author proposes a method based on Pre-training language model. On this basis, taking advantage Bi-directional Long Short-Term Memory (BiLSTM) neural network dealing with wide range contextual information, BERT-BiLSTM-CRF model was constructed. The experimental results tagged...

In neural network binarization, BinaryConnect (BC) and its variants are considered the standard. These methods apply sign function in their forward pass respective gradients backpropagated to update weights. However, derivative of is zero whenever defined, which consequently freezes training. Therefore, implementations BC (e.g., BNN) usually replace backward computation with identity or other approximate gradient alternatives. Although such practice works well empirically, it largely a...

The positioning based on ultra-wideband (UWB) mostly requires the target to carry tags, which is called active positioning. This method a constant power supply for tag and not suitable continuous Therefore, we propose localization passive that achieves position perception without wearing assistive devices. Tomography as one of most promising techniques has many advantages. However, traditional tomography methods require large number fixed-position transceivers form dense links ensure...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

We revisit the efficacy of several practical methods for approximate machine unlearning developed large-scale deep learning. In addition to complying with data deletion requests, one often-cited potential application is remove effects training on poisoned data. experimentally demonstrate that, while existing have been demonstrated be effective in a number evaluation settings (e.g., alleviating membership inference attacks), they fail poisoning, across variety types poisoning attacks...

Availability attacks, or unlearnable examples, are defensive techniques that allow data owners to modify their datasets in ways prevent unauthorized machine learning models from effectively while maintaining the data's intended functionality. It has led release of popular black-box tools for users upload personal and receive protected counterparts. In this work, we show such protections can be substantially bypassed if a small set unprotected in-distribution is available. Specifically, an...