检索结果-内蒙古大学图书馆

19th Pacific Rim International Conference on Artificial Intelligence (PRICAI)

作者： Zhang, Jianming Tao, Jia-Jun Kuang, Li-Dan Gui, Yan Changsha Univ Sci & Technol Sch Comp & Commun Engn Changsha 410114 Peoples R China

ISBN: (纸本)9783031208645;9783031208652

In this paper, we propose a novel key-point detector with only one-level feature with the stride of 8, which is 75.0% less than methods with the stride of 4. Due to the reduction of the feature layers, firstly we adopt a new key-point labeling method, which can make full use of the detection points on the feature map. Secondly, we propose a U-shaped feature fusion module with group residual dense blocks, which works together with grouped convolutional and re-parameterization methods to bring significant improvements while reducing parameters. Thirdly, we use a soft non-key-point branch to re-weight the classification score. Using NVIDIA GeForce 3060 GPU and based on the VOC dataset, the proposed model with repVGG-A0 runs about 51.4% faster than CenterNet with resNet-18, runs 261.3% faster, and achieves higher accuracy than CenterNet with resNet-101 under the resolution of 512 x 512.

关键词： Object detection Key-point detector One-level feature residual dense block re-parameterization

来源：评论

学校读者我要写书评

暂无评论

PreAdapter: Sparse Adaptive Parameter-efficient Transfer Learning for Language Models 7

PreAdapter: Sparse Adaptive Parameter-efficient Transfer Lea...

引用

7th International Conference on Artificial Intelligence and Big Data (ICAIBD)

作者： Mao, Chenyang Jin, Xiaoxiao Yue, Dengfeng Leng, Tuo Shanghai Univ Dept Comp Engn & Sci Shanghai Peoples R China

ISBN: (纸本)9798350385113;9798350385106

As model parameters grow, fine-tuning new models for each task becomes resource-heavy. Parameter-Efficient Transfer Learning (PETL) offers a solution by updating minimal parameters, allowing large pretrained models to adapt to downstream tasks more efficiently. However, existing PETL methods often suffer from increased inference latency. This paper presents 'PreAdapter', an enhancement to traditional adapter structures, overcoming this latency issue. PreAdapter maintains PETL's benefits while avoiding extra inference delay. Specifically, we first demonstrate that adapter modules can be incorporated into large pre-trained language models through structural reparameterization. We explore the optimal deployment sites and the sparse architecture of these adapter modules, aiming to enhance the performance and parameter-efficiency of PreAdapter. To validate the effectiveness of PreAdapter, we perform transfer experiments on the GLUE dataset using the most representative BERT models. Compared to preceding adapter methods, PreAdapter updates 0.6 million fewer parameters in the 110 million parameter BERT-base model, yielding an average performance improvement of 2.7%. In the case of the 335 million parameter BERT-large model, PreAdapter reduces parameter updates by 1.5 million relative to previous methods, leading to an average performance enhancement of 1.7%.

关键词： Adapter re-parameterization Natural Language Processing

来源：评论

学校读者我要写书评

暂无评论

Expectation Thinning Operators Based on Linear Fractional Probability Generating Functions

引用

JOURNAL OF THE INDIAN SOCIETY FOR PROBABILITY AND STATISTICS 2019年第1期20卷 89-107页

作者： Aly, Emad-Eldin A. A. Bouzar, Nadjib Kuwait Univ Dept Stat & OR POB 5969 Safat 13060 Kuwait Univ Indianapolis Dept Math Sci Indianapolis IN 46227 USA

We introduce a two-parameter expectation thinning operator based on a linear fractional probability generating function. The operator is then used to define a first-order integer-valued autoregressive INAR (1) process. Distributional properties of the INAR (1) process are described. We revisit the Bernoulli-geometric INAR (1) process of Bourguignon and Wei ss (Test 26(4):847-868, 2017. https://***/10.1007/s11749-017-0536-4) and we introduce a new stationary INAR (1) process with a compound negative binomial distribution. Lastly, we show how a proper randomization of our operator leads to a generalized notion of monotonicity for distributions on Z+.

关键词： Semigroup INAR (1) process re-parameterization Stationarity Monotonicity

来源：评论

学校读者我要写书评

暂无评论

Estimation of Lower Percentiles Under a Weibull Distribution 1

Estimation of Lower Percentiles Under a Weibull Distribution

引用

1st International Conference on reliability Systems Engineering (ICRSE)

作者： Lv, Shanshan Niu, Zhanwen He, Zhen Qu, Liang Tianjin Univ Coll Management & Econ Tianjin Peoples R China

ISBN: (纸本)9781467385572

We discuss here the estimation of lower percentiles by assuming a Weibull distribution. A log-transformation is made with the Weibull distribution to a smallest extreme value distribution. The location parameter of SEV is re-parameterized by a particular 100pth percentile and the scale parameter of the transformed distribution is assumed to be nonconstant. Statistical inference about the model parameters will be addressed on the basis of the maximum likelihood estimation. And the corresponding confidence intervals of the percentiles are computed based on the normal approximation method. An illustrative example and a simulation study will be presented to show the appropriateness of the method. The simulation study shows that the lower percentile re-parameterization model gives more accurate lower percentile estimates than the commonly used location assumption model.

关键词： re-parameterization nonconstant shape parameter maximum likelihood estimation lower percentiles

来源：评论

学校读者我要写书评

暂无评论

rep-Enhancer: re-parameterizing Neural Network for real-time Low-light Enhancement in Visual Maritime Surveillance 20

Rep-Enhancer: Re-parameterizing Neural Network for Real-time...

引用

20th IEEE International Conference on Embedded and Ubiquitous Computing (EUC)

作者： Li, Xijing Lu, Yuxu Guo, Yu Qu, Jingxiang Liu, Ryan Wen Wuhan Univ Technol Sch Nav Wuhan 430063 Peoples R China Wuhan Univ Technol Chongqing Res Inst Chongqing 401151 Peoples R China

ISBN: (纸本)9798350396355

Vision-based maritime surveillance has become an essential part of the vessel traffic services system. The images collected in low-light maritime conditions often suffer from poor visibility. These images may significantly degenerate the performance of high-level visual tasks and increase the uncertainty in maritime surveillance. To address this problem, we propose a lightweight neural network (rep-Enhancer) for low-light image enhancement. Specifically, we first design a re-parameterizable multi-branch edge extraction module, i.e., spatial domain-oriented convolution block (SDCB). Furthermore, skip connections and spatial attention operations are employed to strengthen the features. By exploiting these well-strengthened edge features, we can enhance the low-light images effectively with the encoder-decoder structure. The experimental results have shown that rep-Enhancer can enhance the low-light image qualifiedly while maintaining great inference efficiency.

关键词： Image enhancement re-parameterization lowlight maritime surveillance edge feature

来源：评论

学校读者我要写书评

暂无评论

Multi-Vision Services Acceleration Framework for IoT Devices

Multi-Vision Services Acceleration Framework for IoT Devices

引用

IEEE International Conference on Web Services (IEEE ICWS)

作者： Xie, Jianhang Ding, Chuntao ren, Shenyuan Li, Yidong Key Lab Big Data Artificial Intelligence Transpor Minist Educ Beijing Peoples R China Beijing Jiaotong Univ Sch Comp Sci & Technol Beijing 100044 Peoples R China

ISBN: (纸本)9798350368567;9798350368550

Deploying a single deep neural network (DNN) model on ubiquitous Internet of Things (IoT) devices to provide multiple vision services (e.g., semantic segmentation service, facial attribute recognition service) has attracted significant interest from industry and academia. However, the most relevant studies have two limitations: (i) the multiple services DNN commonly has low performance due to the gradient interference between different vision services;(ii) the multiple services DNN has low inference speed on IoT devices. To this end, this paper introduces a multiple vision services acceleration framework for IoT devices. The proposed MCBNet contains a novel multi-controllable branching structure to control different services explicitly for improving multiple services performance and a service balance layer with the per-channel trainable masks in specific branches for mitigating gradient interference. In deployment, the proposed framework converts the multi-branch topology MCBNet model into the plain topology rep-MCBNet model by re-parameterization technique for accelerating services inference speed. Experimental results on the CelebA and Cityscapes datasets show that our proposed framework outperforms the state-of-the-art approaches in terms of performance and inference speed. In particular, we deploy the rep-MCBNet on the laptop with Intel mobile processors demonstrating that the proposed framework achieves 1.9x faster processing frames per second than ETR-NLP, with 1.5x fewer parameters and 1.3x fewer floating-point operations.

关键词： Service computing multiple vision services DNN IoT devices re-parameterization

来源：评论

学校读者我要写书评

暂无评论

Estimating m-regimes STAR-GARCH model using QMLE with parameter transformation

Estimating <i>m</i>-regimes STAR-GARCH model using QMLE with...

引用

IMACS World Congress/Modelling and Simulation Society-of-Australia-and-New-Zealand (MSSANZ)/18th MODSIM09 Biennial Conference on Modelling and Simulation

作者： Chan, Felix Theoharakis, Billy Curtin Univ Technol Sch Econ & Finance Bentley WA 6102 Australia

It is well known in the literature that obtaining the parameter estimates for the Smooth Transition Autoregressive-Generalized Autoregressive Conditional Heteroskedasticity (STAR-GARCH) can be problematic due to computational difficulties. Conventional optimization algorithms do not seem to perform well in locating the global optimum of the associated likelihood function. This makes Quasi-Maximum Likelihood Estimator (QMLE) difficult to obtain for STAR-GARCH models in practice. Curiously, there has been very little research investigating the cause of the numerical difficulties in obtaining the parameter estimates for STAR-GARCH using QMLE. The aim of the paper is to investigate the nature of the numerical difficulties using Monte Carlo Simulation. By examining the surface of the log-likelihood function based on simulated data, the results provide several insights into the difficulties in obtaining QMLE for STAR-GARCH models. Based on the findings, the paper also proposes a simple transformation on the parameters to alleviate these difficulties. Monte Carlo simulation results show promising signs for the proposed transform. The asymptotic and robust variance covariance matrices of the original parameter estimates are derived as a function of the transformed parameter estimates, which greatly facilitates inferences on the original parameters. (c) 2010 IMACS. Published by Elsevier B.V. All rights reserved.

关键词： STAR GARCH Monte Carlo simulation re-parameterization

来源：评论

学校读者我要写书评

暂无评论

Kernel Dimension Matters: to Activate Available Kernels for real-time Video Super-resolution 23

Kernel Dimension Matters: to Activate Available Kernels for ...

引用

31st ACM International Conference on Multimedia (MM)

作者： Jin, Shuo Liu, Meiqin Yao, Chao Lin, Chunyu Zhao, Yao Beijing Jiaotong Univ Inst Informat Sci Beijing Key Lab Adv Informat Sci & Network Techno Beijing Peoples R China Univ Sci & Technol Beijing Sch Comp & Commun Engn Beijing Peoples R China

ISBN: (纸本)9798400701085

real-time video super-resolution requires low latency with high-quality reconstruction. Existing methods mostly use pruning schemes or neglect complicated modules to reduce the calculation complexity. However, the video contains large amounts of temporal redundancies due to the inter-frame correlation, which is rarely investigated in existing methods. The static and dynamic information lies in feature maps and represents the redundant complements and temporal offsets respectively. It is crucial to split channels with dynamic and static information for efficient processing. Thus, this paper proposes a kernel-split strategy to activate available kernels for real-time inference. This strategy focuses on the dimensions of convolutional kernels, including the channel and depth dimensions. Available kernel dimensions are activated according to the split of high-value and low-value channels. Specifically, a multi-channel selection unit is designed to discriminate the importance of channels and filter the high-value channels hierarchically. At each hierarchy, low-dimensional convolutional kernels are activated to reuse the low-value channel and re-parameterized convolutional kernels are employed on the high-value channel to merge the depth dimension. In addition, we design a multiple flow deformable alignment module for a sufficient temporal representation with affordable calculation cost. Experimental results demonstrate that our method outperforms other state-of-the-art (SOTA) ones in terms of reconstruction quality and runtime. Codes will be available at https://***/Kimsure/KSNet.

关键词： video super-resolution real-time network re-parameterization kernel split

来源：评论

学校读者我要写书评

暂无评论

real Spike: Learning real-Valued Spikes for Spiking Neural Networks 1

引用

17th European Conference on Computer Vision (ECCV)

作者： Guo, Yufei Zhang, Liwen Chen, Yuanpei Tong, Xinyi Liu, Xiaode Wang, YingLei Huang, Xuhui Ma, Zhe Intelligent Sci & Technol Acad CASIC Beijing 100854 Peoples R China

ISBN: (数字)9783031197758

ISBN: (纸本)9783031197741;9783031197758

Brain-inspired spiking neural networks (SNNs) have recently drawn more and more attention due to their event-driven and energy-efficient characteristics. The integration of storage and computation paradigm on neuromorphic hardwares makes SNNs much different from Deep Neural Networks (DNNs). In this paper, we argue that SNNs may not benefit from the weight-sharing mechanism, which can effectively reduce parameters and improve inference efficiency in DNNs, in some hardwares, and assume that an SNN with unshared convolution kernels could perform better. Motivated by this assumption, a training-inference decoupling method for SNNs named as real Spike is proposed, which not only enjoys both unshared convolution kernels and binary spikes in inference-time but also maintains both shared convolution kernels and real-valued Spikes during training. This decoupling mechanism of SNN is realized by a re-parameterization technique. Furthermore, based on the training-inference-decoupled idea, a series of different forms for implementing real Spike on different levels are presented, which also enjoy shared convolutions in the inference and are friendly to both neuromorphic and non-neuromorphic hardware platforms. A theoretical proof is given to clarify that the real Spike-based SNN network is superior to its vanilla counterpart. Experimental results show that all different real Spike versions can consistently improve the SNN performance. Moreover, the proposed method outperforms the state-of-the-art models on both non-spiking static and neuromorphic datasets.

关键词： Spiking neural network real spike Binary spike Training-inference-decoupled re-parameterization

来源：评论

学校读者我要写书评

暂无评论

rep-MCA-former: An efficient multi-scale convolution attention encoder for text-independent speaker verification

引用

COMPUTER SPEECH AND LANGUAGE 2024年 85卷

作者： Liu, Xiaohu Chen, Defu Wang, Xianbao Xiang, Sheng Zhou, Xuwen Zhejiang Univ Technol Informat Engineer Coll Hangzhou 310023 Zhejiang Peoples R China

In many speaker verification tasks, the quality of speaker embedding is an important factor in affecting speaker verification systems. Advanced speaker embedding extraction networks aim to capture richer speaker features through the multi-branch network architecture. recently, speaker verification systems based on transformer encoders have received much attention, and many satisfactory results have been achieved because transformer encoders can efficiently extract the global features of the speaker (e.g., MFA-Conformer). However, the large number of model parameters and computational latency are common problems faced by the above approaches, which make them difficult to apply to resource-constrained edge terminals. To address this issue, this paper proposes an effective, lightweight transformer model (MCA-former) with multi-scale convolutional self-attention (MCA), which can perform multi-scale modeling and channel modeling in the temporal direction of the input with low computational cost. In addition, in the inference phase of the model, we further develop a systematic re-parameterization method to convert the multi-branch network structure into the single-path topology, effectively improving the inference speed. We investigate the performance of the MCA-former for speaker verification under the VoxCeleb1 test set. The results show that the MCA-based transformer model is more advantageous in terms of the number of parameters and inference efficiency. By applying the re-parameterization, the inference speed of the model is increased by about 30%, and the memory consumption is significantly improved.

关键词： Speaker verification Transformer encoder Multi-scale convolution re-parameterization

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：