检索结果-内蒙古大学图书馆

5th International Conference on High Performance Big Data and intelligent Systems, HDIS 2023

作者： Zhou, Qi Yang, Kai Ke, Zunwang Wang, Gang Zhang, Yugui Jia, Yizhen Cao, Fengcai Ma, Junxiao Liu, Changlin Zhang, Kaijie Wu, Min School of Semiconductor Science and Technology South China Normal University GuangZhou510631 China CGNPC Uranium Industry Development Co. Ltd. Beijing100020 China School of Software Xinjiang University Urumqi830046 China School of Computing and Data Engineering NingboTech University Ningbo315100 China AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China School of Integrated Circuits University of Chinese Academy of Sciences Beijing100049 China Co. Ltd Beijing100176 China School of Information Beijing Forestry University Beijing100083 China

ISBN: (纸本)9798350394160

In real-world scenarios, capturing scenes with excessive dynamic range often leads to the partial loss of highlight or dark area information due to irradiance variations and limitations in the capture capabilities of imaging devices caused by their size and material constraints. This phenomenon gives rise to anomalous exposure problems and triggers the loss of crucial initial details, thereby becoming a significant obstacle to achieving high image quality. Despite numerous researchers’ integration of exposure compensation techniques into the imaging process, the anomalous exposure problem remains unsolved, hindering progress in image processing. Consequently, considerable research is valuable in implementing exposure calibration techniques on visual perception and multimedia devices. This dissertation proposes FARDBUNet, a novel end-to-end exposure calibration model designed to address the anomalous exposure problem. A key innovation of FARDBUNet is the incorporation of a fused attention mechanism, enabling the model to prioritize regions needing enhanced correction while accounting for the global exposure characteristics of the image. Additionally, to tackle the challenge of inadequate information availability in the localized areas due to exposure issues, we introduce an atrous residual dense block that efficiently captures local information by expanding the receptive field. Comprehensive tests validate the effectiveness of our suggested model, which performs on par with leading techniques on expansive exposure datasets. ©2023 IEEE.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Acq: Improving Generative Data-Free Quantization Via Attention Correction

SSRN

引用

SSRN 2023年

作者： Li, Jixing Guo, Xiaozhou Dai, Benzhe Gong, Guoliang Jin, Min Chen, Gang Mao, Wenyu Lu, Huaxiang Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China University of Chinese Academy of Sciences Beijing100089 China Semiconductor Neural Network Intelligent Perception and Computing Technology Beijing Key Laboratory Beijing100083 China Materials and Optoelectronics Research Center University of Chinese Academy of Sciences Beijing100083 China College of Microelectronics University of Chinese Academy of Sciences Beijing China

Data-free quantization aims to achieve model quantization without accessing any authentic sample. It is significant in an application-oriented context involving data privacy. Converting noise vectors into synthetic samples through a generator is a popular data-free quantization method, which is called generative data-free quantization. However, there is a difference in attention between synthetic samples and authentic samples. This is always ignored and restricts the quantization performance. First, since synthetic samples of the same class are prone to have homogenous attention, the quantized network can only learn limited modes of attention. Second, synthetic samples in eval mode and training mode exhibit different attention. Hence, the batch-normalization statistics matching tends to be inaccurate. ACQ is proposed in this paper to fix the attention of synthetic samples. An attention center position-condition generator is established regarding the homogenization of intra-class attention. Restricted by the attention center matching loss, the attention center position is treated as the generator’s condition input to guide synthetic samples in obtaining diverse attention. Moreover, we design adversarial loss of paired synthetic samples under the same condition to prevent the generator from paying overmuch attention to the condition, which may result in mode collapse. To improve the attention similarity of synthetic samples in different network modes, we introduce a consistency penalty to guarantee accurate BN statistics matching. The experimental results demonstrate that ACQ effectively improves the attention problems of synthetic samples. Under various training settings, ACQ achieves the best quantization performance. For the 4-bit quantization of Resnet18 and Resnet50, ACQ reaches 67.55% and 72.23% accuracy, respectively. © 2023, The Authors. All rights reserved.

关键词： Data privacy

来源：评论

学校读者我要写书评

暂无评论

DoctorGPT: A Large Language Model with Chinese Medical Question-Answering Capabilities

DoctorGPT: A Large Language Model with Chinese Medical Quest...

引用

High Performance Big Data and intelligent Systems (HPBD&IS)

作者： Wenqiang Li Lina Yu Min Wu Jingyi Liu Meilan Hao Yanjie Li AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing China School of Electronic Electrical and Communication Engineering & School of Integrated Circuits University of Chinese Academy of Sciences Beijing China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing China

Large Language Models (LLMs) have made incredible strides recently in understanding and reacting to user intents. However, these models typically excel in English and have not been specifically trained for medical applications, leading to suboptimal performance in responding to medical inquiries such as diagnostic queries and drug recommendations. In this paper, we propose DoctorGPT, a domain-specific large language model tailored for medical question-answering tasks. DoctorGPT leverages the open-source Baichuan2 as its foundational model, undergoes extensive pre-training on medical encyclopedic data to incorporate medical knowledge, and subsequently undergoes fine-tuning on a dataset consisting of two million medical instruction-dialogue pairs to enhance its question-answering capabilities. When compared to general-purpose large models, DoctorGPT demonstrates significant advantages in Chinese medical question-answerinz (O&A) tasks.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A neural-guided dynamic symbolic network for exploring mathematical expressions from data 24

A neural-guided dynamic symbolic network for exploring mathe...

引用

Proceedings of the 41st International Conference on Machine Learning

作者： Wenqiang Li Weijun Li Lina Yu Min Wu Linjun Sun Jingyi Liu Yanjie Li Shu Wei Yusong Deng Meilan Hao AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing China and School of Electronic Electrical and Communication Engineering & School of Integrated Circuits University of Chinese Academy of Sciences Beijing China and Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing China AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing China and School of Electronic Electrical and Communication Engineering & School of Integrated Circuits and Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing China and Center of Materials Science and Optoelectronics Engineering University of Chinese Academy of Sciences Beijing China

Symbolic regression (SR) is a powerful technique for discovering the underlying mathematical expressions from observed data. Inspired by the success of deep learning, recent deep generative SR methods have shown promising results. However, these methods face difficulties in processing high-dimensional problems and learning constants due to the large search space, and they don't scale well to unseen problems. In this work, we propose DYSYMNET, a novel neural-guided Dynamic Symbolic network for SR. Instead of searching for expressions within a large search space, we explore symbolic networks with various structures, guided by reinforcement learning, and optimize them to identify expressions that better-fitting the data. Based on extensive numerical experiments on low-dimensional public standard benchmarks and the well-known SRBench with more variables, DYSYMNET shows clear superiority over several representative baseline models. Open source code is available at https://***/AILWQ/DySymNet.

关键词：

来源：评论

学校读者我要写书评

暂无评论

ACQ: Improving Generative Data-free Quantization Via Attention Correction

arXiv

引用

arXiv 2023年

作者： Li, Jixing Guo, Xiaozhou Dai, Benzhe Gong, Guoliang Jin, Min Chen, Gang Mao, Wenyu Lu, Huaxiang Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China University of Chinese Academy of Sciences Beijing100049 China Semiconductor Neural Network Intelligent Perception and Computing Technology Beijing Key Laboratory Beijing100083 China Materials and Optoelectronics Research Center University of Chinese Academy of Sciences Beijing100049 China College of Microelectronics University of Chinese Academy of Sciences Beijing100049 China

Data-free quantization aims to achieve model quantization without accessing any authentic sample. It is significant in an application-oriented context involving data privacy. Converting noise vectors into synthetic samples through a generator is a popular data-free quantization method, which is called generative data-free quantization. However, there is a difference in attention between synthetic samples and authentic samples. This is always ignored and restricts the quantization performance. First, since synthetic samples of the same class are prone to have homogenous attention, the quantized network can only learn limited modes of attention. Second, synthetic samples in eval mode and training mode exhibit different attention. Hence, the batch-normalization statistics matching tends to be inaccurate. ACQ is proposed in this paper to fix the attention of synthetic samples. An attention center position-condition generator is established regarding the homogenization of intra-class attention. Restricted by the attention center matching loss, the attention center position is treated as the generator's condition input to guide synthetic samples in obtaining diverse attention. Moreover, we design adversarial loss of paired synthetic samples under the same condition to prevent the generator from paying overmuch attention to the condition, which may result in mode collapse. To improve the attention similarity of synthetic samples in different network modes, we introduce a consistency penalty to guarantee accurate BN statistics matching. The experimental results demonstrate that ACQ effectively improves the attention problems of synthetic samples. Under various training settings, ACQ achieves the best quantization performance. For the 4-bit quantization of Resnet18 and Resnet50, ACQ reaches 67.55% and 72.23% accuracy, respectively. Copyright © 2023, The Authors. All rights reserved.

关键词： Data privacy

来源：评论

学校读者我要写书评

暂无评论

GmFace: A mathematical model for face image representation using multi-gaussian

arXiv

引用

arXiv 2020年

作者： Zhang, Liping Li, Weijun Yu, Lina Dong, Xiaoli Sun, Linjun Ning, Xin Xu, Jian Qin, Hong Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Center of Materials Science and Optoelectronics Engineering School of Microelectronics University of Chinese Academy of Sciences Beijing100049 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China

Establishing mathematical models is a ubiquitous and effective method to understand the objective world. Due to complex physiological structures and dynamic behaviors, mathematical representation of the human face is an especially challenging task. A mathematical model for face image representation called GmFace is proposed in the form of a multi-Gaussian function in this paper. The model utilizes the advantages of two-dimensional Gaussian function which provides a symmetric bell surface with a shape that can be controlled by parameters. The GmNet is then designed using Gaussian functions as neurons, with parameters that correspond to each of the parameters of GmFace in order to transform the problem of GmFace parameter solving into a network optimization problem of GmNet. The face modeling process can be described by the following steps: (1) GmNet initialization;(2) feeding GmNet with face image(s);(3) training GmNet until convergence;(4) drawing out the parameters of GmNet (as the same as GmFace);(5) recording the face model GmFace. Furthermore, using GmFace, several face image transformation operations can be realized mathematically through simple parameter computation. Copyright © 2020, The Authors. All rights reserved.

关键词： Gaussian distribution

来源：评论

学校读者我要写书评

暂无评论

Rec-Symnet: Symbolic network-Based Rectifiable Learning Framework for Symbolic Regression

SSRN

引用

SSRN 2023年

作者： Liu, Jingyi Li, Weijun Yu, Lina Wu, Min Sun, Linjun Li, Wenqiang Li, Yanjie Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Center of Materials Science and Optoelectronics Engineering School of Integrated Circuits University of Chinese Academy of Sciences Beijing100049 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China

Symbolic regression (SR) can be utilized to unveil the underlying mathematical expressions that describe a given set of observed data. At present, SR can be categorized into two methods: learning-from-scratch and learning-with-experience. Compared to learning-from-scratch, learning-with-experience yields results that are comparable to those of several benchmarks and incurs significantly lower time costs for obtaining expressions. However, the learning-with-experience model performs poorly in terms of unseen data distributions and lacks a rectification tool, apart from constant optimization, which exhibits limited performance. In this study, we propose a Symbolic network-based Rectifiable Learning Framework (Rec-SymNet) that possesses the ability to correct errors. Rec-SymNet adopts Symbolic network (SymNet) to represent an expression, and the encoding of SymNet is designed to provide supervised information, with numerous self-generated expressions, to train a policy net (PolicyNet). The training of PolicyNet can offer prior knowledge to guide effective searches. Subsequently, the incorrectly predicted expressions are revised via a rectification mechanism. This rectification mechanism endows Rec-SymNet with broader applicability. Experimental results demonstrate that our proposed method achieves the highest averaged coefficient of determination on self-generated datasets when compared with other state-of-the-art methods and yields more accurate results in public datasets. © 2023, The Authors. All rights reserved.

关键词： Regression analysis

来源：评论

学校读者我要写书评

暂无评论

PruneSymNet: A Symbolic neural network and Pruning Algorithm for Symbolic Regression

arXiv

引用

arXiv 2024年

作者： Wu, Min Li, Weijun Yu, Lina Li, Wenqiang Liu, Jingyi Li, Yanjie Hao, Meilan AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Center of Materials Science and Optoelectronics Engineering School of Microelectronics University of Chinese Academy of Sciences Beijing100049 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China

Symbolic regression aims to derive interpretable symbolic expressions from data in order to better understand and interpret data. In this study, a symbolic network called PruneSymNet is proposed for symbolic regression. This is a novel neural network whose activation function consists of common elementary functions and operators. The whole network is differentiable and can be trained by gradient descent method. Each subnetwork in the network corresponds to an expression, and our goal is to extract such subnetworks to get the desired symbolic expression. Therefore, a greedy pruning algorithm is proposed to prune the network into a subnetwork while ensuring the accuracy of data fitting. The proposed greedy pruning algorithm preserves the edge with the least loss in each pruning, but greedy algorithm often can not get the optimal solution. In order to alleviate this problem, we combine beam search during pruning to obtain multiple candidate expressions each time, and finally select the expression with the smallest loss as the final result. It was tested on the public data set and compared with the current popular algorithms. The results showed that the proposed algorithm had better accuracy. Copyright © 2024, The Authors. All rights reserved.

关键词： Gradient methods

来源：评论

学校读者我要写书评

暂无评论

Learning continuous face representation with explicit functions

arXiv

引用

arXiv 2021年

作者： Zhang, Liping Li, Weijun Sun, Linjun Yu, Lina Ning, Xin Dong, Xiaoli Xu, Jian Qin, Hong Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Center of Materials Science and Optoelectronics Engineering School of Microelectronics University of Chinese Academy of Sciences Beijing100049 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Institute of Semiconductors Beijing100083 China Cognitive Computing Technology Joint Laboratory Wave Group Beijing100083 China

How to represent a face pattern? While it is presented in a continuous way in our visual system, computers often store and process the face image in a discrete manner with 2D arrays of pixels. In this study, we attempt to learn a continuous representation for face images with explicit functions. First, we propose an explicit model (EmFace) for human face representation in the form of a finite sum of mathematical terms, where each term is an analytic function element. Further, to estimate the unknown parameters of EmFace, a novel neural network, EmNet, is designed with an encoder-decoder structure and trained using the backpropagation algorithm, where the encoder is defined by a deep convolutional neural network and the decoder is an explicit mathematical expression of EmFace. Experimental results show that EmFace has a higher representation performance on faces with various expressions, postures, and other factors, compared to that of other methods. Furthermore, EmFace achieves reasonable performance on several face image processing tasks, including face image restoration, denoising, and transformation. Copyright © 2021, The Authors. All rights reserved.

关键词： Functions

来源：评论

学校读者我要写书评

暂无评论

Incorporating Higher-Knowledge into Deep Symbolic Regression Under Generative neural network Framework

SSRN

引用

SSRN 2024年

作者： Liu, Jingyi Li, Weijun Yu, Lina Wu, Min Li, Wenqiang Li, Yanjie Hao, Meilan Annlab Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Center of Materials Science and Optoelectronics Engineering School of Integrated Circuits University of Chinese Academy of Sciences Beijing100049 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China

This paper introduces MSR, a generative neural network-based framework that incorporates an automatic module learning feature, aiming to enhance the search accuracy of Symbolic Regression (SR). Unlike existing deep model-based SR approaches, MSR identifies valuable sub-structures named modules, which are utilized to construct the final solution. Modules are considered high-order knowledge and act as fundamental operators, expanding the search library of SR. The proposed algorithm enables self-learning or self-evolution of modules as part of the learning component. By incorporating variables or partial solutions as input arguments, MSR facilitates the formation of sophisticated, higher-order, and constructive sub-structures, thereby improving search accuracy. Specifically, we propose a module extraction strategy that generates modules hierarchically from the expression tree, along with a module update mechanism designed to effectively eliminate unnecessary modules while incorporating new useful ones. To validate the effectiveness of MSR, we conducted experiments on 60 SR problems. Through ablation studies, we analyze the impact of module utilization and the module update mechanism. The experimental results demonstrate that our method outperforms scenarios where modules are not utilized or random updates are applied. Furthermore, our approach exhibits higher accuracy in SR problems involving constants, surpassing some state-of-the-arts. Our code has been released at https://***/r/MSR. © 2024, The Authors. All rights reserved.

关键词： Extraction

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：