检索结果-内蒙古大学图书馆

arXiv 2021年

作者： Hou, Liang Cao, Qi Shen, Huawei Pan, Siyuan Li, Xiaoshuang Cheng, Xueqi Data Intelligence System Research Center Institute of Computing Technology Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Beijing China Shanghai Jiao Tong University Shanghai China CAS Key Laboratory of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences Beijing China

Conditional generative models aim to learn the underlying joint distribution of data and labels to achieve conditional data generation. Among them, the auxiliary classifier generative adversarial network (AC-GAN) has been widely used, but suffers from the problem of low intra-class diversity of the generated samples. The fundamental reason pointed out in this paper is that the classifier of AC-GAN is generator-agnostic, which therefore cannot provide informative guidance for the generator to approach the joint distribution, resulting in a minimization of the conditional entropy that decreases the intra-class diversity. Motivated by this understanding, we propose a novel conditional GAN with an auxiliary discriminative classifier (ADC-GAN) to resolve the above problem. Specifically, the proposed auxiliary discriminative classifier becomes generator-aware by recognizing the class-labels of the real data and the generated data discriminatively. Our theoretical analysis reveals that the generator can faithfully learn the joint distribution even without the original discriminator, making the proposed ADC-GAN robust to the value of the coefficient hyperparameter and the selection of the GAN loss, and stable during training. Extensive experimental results on synthetic and real-world datasets demonstrate the superiority of ADC-GAN in conditional generative modeling compared to state-of-the-art classifier-based and projection-based conditional GANs. Copyright © 2021, The Authors. All rights reserved.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

ADFQ-ViT: Activation-Distribution-Friendly Post-Training Quantization for Vision Transformers

arXiv

引用

arXiv 2024年

作者： Jiang, Yanfeng Sun, Ning Xie, Xueshuo Yang, Fei Li, Tao College of Computer Science Nankai University Tianjin China Tianjin Key Laboratory of Network and Data Security Technology Tianjin China Zhejiang Lab Zhejiang Hangzhou China Haihe Lab of ITAI Tianjin China

Vision Transformers (ViTs) have exhibited exceptional performance across diverse computer vision tasks, while their substantial parameter size incurs significantly increased memory and computational demands, impeding effective inference on resource-constrained devices. Quantization has emerged as a promising solution to mitigate these challenges, yet existing methods still suffer from significant accuracy loss at low-bit. We attribute this issue to the distinctive distributions of post-LayerNorm and post-GELU activations within ViTs, rendering conventional hardware-friendly quantizers ineffective, particularly in low-bit scenarios. To address this issue, we propose a novel framework called Activation-Distribution-Friendly post-training Quantization for Vision Transformers, ADFQ-ViT. Concretely, we introduce the Per-Patch Outlier-aware Quantizer to tackle irregular outliers in post-LayerNorm activations. This quantizer refines the granularity of the uniform quantizer to a per-patch level while retaining a minimal subset of values exceeding a threshold at full-precision. To handle the non-uniform distributions of post-GELU activations between positive and negative regions, we design the Shift-Log2 Quantizer, which shifts all elements to the positive region and then applies log2 quantization. Moreover, we present the Attention-score enhanced Module-wise Optimization which adjusts the parameters of each quantizer by reconstructing errors to further mitigate quantization error. Extensive experiments demonstrate ADFQ-ViT provides significant improvements over various baselines in image classification, object detection, and instance segmentation tasks at 4-bit. Specifically, when quantizing the ViT-B model to 4-bit, we achieve a 10.23% improvement in Top-1 accuracy on the ImageNet dataset. Copyright © 2024, The Authors. All rights reserved.

关键词： Distribution transformers

来源：评论

学校读者我要写书评

暂无评论

Balanced distribution adaptation for transfer learning

arXiv

引用

arXiv 2018年

作者： Wang, Jindong Chen, Yiqiang Hao, Shuji Feng, Wenjie Shenk, Zhiqi Beijing Key Laboratory of Mobile Computing and Pervasive Device CAS Key Laboratory of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Institute of High Performance Computing A*STAR School of Computer Science and Engineering Nanyang Technological University Singapore Singapore

Transfer learning has achieved promising results by leveraging knowledge from the source domain to annotate the target domain which has few or none labels. Existing methods often seek to minimize the distribution divergence between domains, such as the marginal distribution, the conditional distribution or both. However, these two distances are often treated equally in existing algorithms, which will result in poor performance in real applications. Moreover, existing methods usually assume that the dataset is balanced, which also limits their performances on imbalanced tasks that are quite common in real problems. To tackle the distribution adaptation problem, in this paper, we propose a novel transfer learning approach, named as Balanced Distribution Adaptation (BDA), which can adaptively leverage the importance of the marginal and conditional distribution discrepancies, and several existing methods can be treated as special cases of BDA. Based on BDA, we also propose a novel Weighted Balanced Distribution Adaptation (W-BDA) algorithm to tackle the class imbalance issue in transfer learning. W-BDA not only considers the distribution adaptation between domains but also adaptively changes the weight of each class. To evaluate the proposed methods, we conduct extensive experiments on several transfer learning tasks, which demonstrate the effectiveness of our proposed algorithms over several state-of-the-art methods. Copyright © 2018, The Authors. All rights reserved.

关键词： Learning algorithms

来源：评论

学校读者我要写书评

暂无评论

Adfq-Vit: Activation-Distribution-Friendly Post-Training Quantization for Vision Transformers

SSRN

引用

SSRN 2024年

关键词： Chemical activation

来源：评论

学校读者我要写书评

暂无评论

Time-Frequency Analysis-Based Transient Harmonic Feature Extraction for Load Monitoring

Time-Frequency Analysis-Based Transient Harmonic Feature Ext...

引用

International Conference on Parallel and Distributed Systems (ICPADS)

作者： Peng Xia Hao Zhou Shenyao Jiang Fan Deng Zhi Liu Xiang-Yang Li LINKE Lab School of Computer Science and Technology University of Science and Technology of China CAS Key Laboratory of Wireless-Optical Communications University of Science and Technology of China Deqing Alpha Innovation Institute Huzhou Zhejiang China Department of Computer and Network Engineering The University of Electro-Communications Tokyo Japan

Feature extraction is important for non-intrusive appliance load monitoring (NILM), because it includes crucial steps which altogether transform raw signals into distinct features of appliances. Although numerous studies have achieved good results using transient harmonic features, their performances degrade in practical scenarios. In this paper, a time-frequency analysis-based framework for transient harmonic feature extraction is proposed, which takes all factors that would eventually influence NILM system into account. More specffically, the framework proposes a new qualitative transient harmonic feature, along with current retrieval and data augmentation methods. The proposed framework is evaluated on two well-known datasets (i.e., Controlled On/Off Loads Library and Home Equipment laboratory dataset). As compared with the other state-of-art methods, the proposed framework achieves similar accuracy (i.e., more than 96%) on single-load data, and more than 33% accuracy improvement over multi-load data (e.g., from 59.4% to 93% when six devices are active at the same time).

关键词： Load monitoring Home appliances Time-frequency analysis Transforms Harmonic analysis Feature extraction Libraries

来源：评论

学校读者我要写书评

暂无评论

A reduced collatz dynamics maps to a residue class, and its count of x/2 over count of 3*x+1 is larger than ln3/ln2

TechRxiv

引用

TechRxiv 2020年

作者： Ren, Wei School of Computer Science China University of Geosciences Wuhan China Key Laboratory of Network Assessment Technology CAS [Institute of Information Engineering Chinese Academy of Sciences Beijing100093 China Guizhou Provincial Key Laboratory Public Big Data Guizhou University Guizhou China

We propose Reduced Collatz conjecture and prove that it is equivalent to Collatz conjecture but more primitive due to reduced dynamics. We study reduced dynamics (that consists of occurred computations from any starting integer to the first integer less than it), because it is the component of original dynamics (from any starting integer to 1). Reduced dynamics is denoted as a sequence of "I" that represents (3*x+1)/2 and "O" that represents x/2. Here 3*x+1 and x/2 are combined together because 3*x+1 is always even and thus followed by x/2. We discover and prove two key properties on reduced dynamics: (1) Reduced dynamics is invertible. That is, given a reduced dynamics, a residue class that presents such reduced dynamics, can be computed directly by our derived formula. (2) Reduced dynamics can be constructed algorithmically, instead of by computing concrete 3*x+1 and x/2 step by step. We discover the sufficient and necessary condition that guarantees a sequence consisting of "I" and "O" to be a reduced dynamics. Counting from the beginning of a sequence, if and only if the count of x/2 over the count of 3*x+1 is larger than ln3/ln2, reduced dynamics will be obtained (i.e., current integer will be less than starting integer). © 2020, CC BY.

关键词： Dynamics

来源：评论

学校读者我要写书评

暂无评论

Reduced collatz dynamics is periodical and the period equals 2 to the power of the count of x/2

TechRxiv

引用

TechRxiv 2020年

作者： Ren, Wei School of Computer Science China University of Geosciences Wuhan China Key Laboratory of Network Assessment Technology CAS Institute of Information Engineering Chinese Academy of Sciences Beijing100093 China Guizhou Provincial Key Laboratory of Public Big Data Guizhou University Guizhou China

We propose Reduced Collatz Conjecture that is equivalent to Collatz Conjecture but is easier to explore, because reduced dynamics is more primitive than original dynamics and presents better structures (e.g., period and ratio). Reduced dynamics (that are occurred computation sequence from a starting integer to the first integer less than the starting integer) is the component of original dynamics (from a starting integer to 1). Reduced dynamics of x is represented by a sequence of computation that is either (3*x+1)/2 or x/2, because 3*x+1 is always even and followed by x/2. We prove that reduced dynamics is periodical and its period equals 2 to the power of the count of x/2. More specifically, if there exists reduced dynamics of x, then there exists reduced dynamics of x+P, where P equals 2L and L is the total count of x/2 computations in reduced dynamics of x (equivalently, L is the length of the sequence). © 2020, CC BY.

关键词： Dynamics

来源：评论

学校读者我要写书评

暂无评论

Collatz dynamics is partitioned by residue class regularly

TechRxiv

引用

TechRxiv 2020年

We propose Reduced Collatz Conjecture that is equivalent to Collatz Conjecture, which states that every positive integer can return to an integer less than it, instead of 1. Reduced Collatz Conjecture should be easier because some properties are presented in reduced dynamics, rather than in original dynamics (e.g., ratio and period). Reduced dynamics is a computation sequence from starting integer to the first integer less than it, and original dynamics is a computation sequence from starting integer to 1. Reduced dynamics is a component of original dynamics. We denote dynamics of x as a sequence of either computations in terms of "I" that represents (3*x+1)/2 and "O" that represents x/2. Here 3*x+1 and x/2 are combined together, because 3*x+1 is always even and followed by x/2. We formally prove that all positive integers are partitioned into two halves and either presents "I" or "O" in next ongoing computation. More specifically, (1) if any positive integer x that is i module 2t (i is an odd integer) is given, then the first t computations (each one is either "I" or "O" corresponding to whether current integer is odd or even) will be identical with that of i. (2) If current integer after t computations (in terms of "I" or "O") is less than x, then reduced dynamics of x is available. Otherwise, the residue class of x (namely, i module 2t) can be partitioned into two halves (namely, i module 2t+1 and i+2t module 2t+1), and either half presents "I" or "O" in intermediately forthcoming (t+1)-th computation. © 2020, CC BY.

关键词： Dynamics

来源：评论

学校读者我要写书评

暂无评论

The minority matters: a diversity-promoting collaborative metric learning algorithm 22

The minority matters: a diversity-promoting collaborative me...

引用

Proceedings of the 36th International Conference on Neural Information Processing Systems

作者： Shilong Bao Qianqian Xu Zhiyong Yang Yuan He Xiaochun Cao Qingming Huang State Key Laboratory of Information Security Institute of Information Engineering CAS and School of Cyber Security University of Chinese Academy of Sciences Key Lab. of Intelligent Information Processing Institute of Computing Technology CAS School of Computer Science and Tech. University of Chinese Academy of Sciences Alibaba Group School of Cyber Science and Technology Shenzhen Campus Sun Yat-sen University Key Lab. of Intelligent Information Processing Institute of Computing Technology CAS and School of Computer Science and Tech. University of Chinese Academy of Sciences and Key Laboratory of Big Data Mining and Knowledge Management CAS and Peng Cheng Laboratory

ISBN: (纸本)9781713871088

Collaborative Metric Learning (CML) has recently emerged as a popular method in recommendation systems (RS), closing the gap between metric learning and Collaborative Filtering. Following the convention of RS, existing methods exploit unique user representation in their model design. This paper focuses on a challenging scenario where a user has multiple categories of interests. Under this setting, we argue that the unique user representation might induce preference bias, especially when the item category distribution is imbalanced. To address this issue, we propose a novel method called Diversity-Promoting Collaborative Metric Learning (DPCML), with the hope of considering the commonly ignored minority interest of the user. The key idea behind DPCML is to include a multiple set of representations for each user in the system. Based on this embedding paradigm, user preference toward an item is aggregated from different embeddings by taking the minimum item-user distance among the user embedding set. Furthermore, we observe that the diversity of the embeddings for the same user also plays an essential role in the model. To this end, we propose a Diversity Control Regularization Scheme (DCRS) to accommodate the multi-vector representation strategy better. Theoretically, we show that DPCML could generalize well to unseen test data by tackling the challenge of the annoying operation that comes from the minimum value. Experiments over a range of benchmark datasets speak to the efficacy of DPCML.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Proof-of-Principle Demonstration of Passive Decoy-State Quantum Digital Signatures Over 200 km

引用

Physical Review Applied 2018年第3期10卷 034033-034033页

作者： Chun-Hui Zhang Xing-Yu Zhou Hua-Jian Ding Chun-Mei Zhang Guang-Can Guo Qin Wang Institute of Quantum Information and Technology Nanjing University of Posts and Telecommunications Nanjing 210003 China Broadband Wireless Communication and Senser Network Technology Key Lab of Ministry of Education NUPT Nanjing 210003 China Telecommunication and Networks National Engineering Research Center NUPT Nanjing 210003 China Key Laboratory of Quantum Information CAS University of Science and Technology of China Hefei Anhui 230026 China

Quantum digital signatures (QDSs) offer the information-theoretic security verified by quantum mechanics to send a message that is robust against forging and repudiation. We report a proof-of-principle demonstration of a passive decoy-state QDS system that can avoid leaking the modulation information of decoy states. To realize this, four local detection events are generated from the idler light of a parametric down-conversion process, and are then used as decoy states to estimate channel parameters more accurately. As a result, we successfully sign a more-than-one-bit message within 7 s at 100 km. Besides, we achieve a transmission distance of 200 km, which represents a further step toward the practical application of QDSs.

关键词： Photon pairs & parametric down-conversion Quantum communication Quantum cryptography

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：