Previous works utilized "smaller-norm-less-important" criterion to prune filters with smaller norm values in a convolutional neural network. In this paper, we analyze norm-based and point out that its effectiveness depends on two requirements are not always met: (1) the deviation of should be large; (2) minimum small. To solve problem, propose novel filter pruning method, namely Filter Pruning via Geometric Median (FPGM), compress model regardless those requirements. Unlike previous methods,...

This paper proposed a Soft Filter Pruning (SFP) method to accelerate the inference procedure of deep Convolutional Neural Networks (CNNs). Specifically, SFP enables pruned filters be updated when training model after pruning. has two advantages over previous works: (1) Larger capacity. Updating previously provides our approach with larger optimization space than fixing zero. Therefore, network trained by capacity learn from data. (2) Less dependence on pretrained model. Large train scratch...

Filter pruning has been widely applied to neural network compression and acceleration. Existing methods usually utilize pre-defined criteria, such as Lp-norm, prune unimportant filters. There are two major limitations these methods. First, existing fail consider the variety of filter distribution across layers. To extract features coarse level fine level, filters different layers have various distributions. Therefore, it is not suitable same criteria functional Second, prevailing...

Deeper and wider convolutional neural networks (CNNs) achieve superior performance but bring expensive computation cost. Accelerating such overparameterized network has received increased attention. A typical pruning algorithm is a three-stage pipeline, i.e., training, pruning, retraining. Prevailing approaches fix the pruned filters to zero during retraining and, thus, significantly reduce optimization space. Besides, they directly prune large number of at first, which would cause...

The remarkable performance of deep Convolutional neural networks (CNNs) is generally attributed to their deeper and wider architectures, which can come with significant computational costs. Pruning has thus gained interest since it effectively lowers storage In contrast weight pruning, results in unstructured models, structured pruning provides the benefit realistic acceleration by producing models that are friendly hardware implementation. special requirements have led discovery numerous...

This paper focuses on network pruning for image retrieval acceleration. Prevailing works target at the discriminative feature learning, while little attention is paid to how accelerate model inference, which should be taken into consideration in real-world practice. The challenge of models that middle-level preserved as much possible. Such different requirements and classification make traditional methods not suitable our task. To solve problem, we propose a new Progressive Local Filter...

Filter pruning is effective to reduce the computational costs of neural networks. Existing methods show that updating previous pruned filter would enable large model capacity and achieve better performance. However, during iterative process, even if network weights are updated new values, criterion remains same. In addition, when evaluating importance, only magnitude information filters considered. in networks, do not work individually, but they affect other filters. As a result, each...

Physics-informed neural networks (PINNs) have emerged as a significant endeavor in recent years to utilize artificial intelligence technology for solving various partial differential equations (PDEs). Nevertheless, the vanilla PINN model structure encounters challenges accurately approximating solutions at hard-to-fit regions with, instance, "stiffness" points characterized by fast-paced alterations timescale. To this end, we introduce novel architecture based on PINN, named loss-attentional...

This paper proposed a Soft Filter Pruning (SFP) method to accelerate the inference procedure of deep Convolutional Neural Networks (CNNs). Specifically, SFP enables pruned filters be updated when training model after pruning. has two advantages over previous works: (1) Larger capacity. Updating previously provides our approach with larger optimization space than fixing zero. Therefore, network trained by capacity learn from data. (2) Less dependence on pre-trained model. Large train scratch...

Deploying machine learning (ML)-based intrusion detection systems (IDS) is an effective way to improve the security of industrial control (ICS). However, ML models themselves are vulnerable adversarial examples, generated by deliberately adding subtle perturbation input sample that some people not aware of, causing model give a false output with high confidence. In this article, our goal investigate possibility stealthy cyber attacks towards IDS, including injection attack, function code...

With the rapid growth of smart terminals in recent years, crowdsensing which utilizes human intelligence to solve complicated problems have gained considerable interest and exploit. The majority existing systems rely on a trusted third-party platform complete sensing tasks collect large-scale data. However, cannot completely ensure trust real world. issues security privacy caused by center should not be ignored. In this paper, we propose decentralized privacy-preserving model based twice...

Deep learning has shown significant successes in person reidentification (re-id) tasks. However, most existing works focus on discriminative feature and impose complex neural networks, suffering from low inference efficiency. In fact, extraction time is also crucial for real-world applications lightweight models are needed. Prevailing pruning methods usually pay attention to compact classification models. these suboptimal compacting re-id models, which produce continuous features sensitive...

Predicting the mechanical properties of hot‐rolled strip poses significant challenges due to intricate interplay multi‐dimensional similarities within sample analysis and time‐varying characteristics actual production data. Relying on single similarity metrics select appropriate samples becomes inadequate, hindering timely accurate predictions. To address these issues, in this article, a new approach is proposed predict based combining multi‐dimensional‐feature‐weighted (MDFWS) integrated...

Low-field nuclear magnetic resonance (NMR) acts as an indispensable borehole logging method for the pore size characterization and formation evaluation, including estimation of reservoir parameters fluid discrimination.The YGH basin , located in northern South China Sea, is characterized by rapid subsidence rate, high geothermal gradient, pressure coefficient. The overpressure low-permeability reservoirs Basin featured with Fine-grained lithology, elevated shale content, minute sizes,...

Currently, there is insufficient confidence in the application of soft rock major projects, leading to overly conservative design foundation bearing capacity. Clarifying stress distribution and failure process with depth offer guiding significant for determining capacity practical geotechnical engineering. Based on arch project Fifth Bridge Lantian Yangtze River, this paper investigated foundations its correction rule by analyzing field test results numerical simulations.  A formula...

With the popularization of intelligent terminals, crowdsensing has become increasingly prominent because its advantages, such as low cost, high convenience, and fast speed in conducting tasks. However, quality data collected through is varied difficult to evaluate. Furthermore, existing control methods are mostly based on a central platform, which not completely trusted reality results existence fraud other problems. To solve these two questions, model two-consensus blockchain proposed this...

Pedestrian tracking is an important aspect of autonomous vehicles environment perception in a vehicle running environment. The performance the existing pedestrian algorithms limited by complex traffic environment, changeable appearance characteristics pedestrians and frequent occlusion interaction, which leads to insufficient accuracy stability tracking. Therefore, this paper proposes detector–tracker integration framework for Firstly, objects detector based on improved YOLOv7 network was...

We propose a general Variational Embedding Learning Framework (VELF) for alleviating the severe cold-start problem in CTR prediction. VELF addresses cold start via over-fits caused by data-sparsity two ways: learning probabilistic embedding, and incorporating trainable regularized priors which utilize rich side information of users advertisements (Ads). The techniques are naturally integrated into variational inference framework, forming an end-to-end training process. Abundant empirical...

Unsupervised graph-structure learning (GSL) which aims to learn an effective graph structure applied arbitrary downstream tasks by data itself without any labels' guidance, has recently received increasing attention in various real applications. Although several existing unsupervised GSL achieved superior performance different analytical tasks, how utilize the popular masked autoencoder sufficiently acquire supervision information from for improving effectiveness of learned been not...

This article reported the mechanism of Anlotinib in gastric cancer treatment. Gastric cells were treated with (8 μM) and transfected by STING shRNA vectors. Cell counting kit-8 assay, wounding healing Transwell experiment applied for proliferation, migration, invasion detection. PD-L1 fluorescence intensity was explored flow cytometry. IFN-β level researched enzyme-linked immunosorbent reaction. Xenograft tumor performed administering mice anti-PD-L1 antibody. Immunohistochemistry western...

Previous works utilized ''smaller-norm-less-important'' criterion to prune filters with smaller norm values in a convolutional neural network. In this paper, we analyze norm-based and point out that its effectiveness depends on two requirements are not always met: (1) the deviation of should be large; (2) minimum small. To solve problem, propose novel filter pruning method, namely Filter Pruning via Geometric Median (FPGM), compress model regardless those requirements. Unlike previous...

The longstanding goals of federated learning (FL) require rigorous privacy guarantees and low communication overhead while holding a relatively high model accuracy. However, simultaneously achieving all the is extremely challenging. In this paper, we propose novel framework called hierarchical (H-FL) to tackle challenge. Considering degradation performance due statistic heterogeneity training data, devise runtime distribution reconstruction strategy, which reallocates clients appropriately...

Improving the performance of click-through rate (CTR) prediction remains one core tasks in online advertising systems. With rise deep learning, CTR models with networks remarkably enhance model capacities. In models, exploiting users' historical data is essential for learning behaviors and interests. As existing works neglect importance temporal signals when embed clicking records, we propose a time-aware attention which explicitly uses absolute expressing periodic relative relation between...

Conversion rate (CVR) prediction is becoming increasingly important in the multi-billion dollar online display advertising industry. It has two major challenges: firstly, scarce user history data very complicated and non-linear; secondly, time delay between clicks corresponding conversions can be large, e.g., ranging from seconds to weeks. Existing models usually suffer such delayed conversion behaviors. In this paper, we propose a novel deep learning framework tackle challenges....