检索结果-内蒙古大学图书馆

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Zhou, Baojian Sun, Yifan Harikandeh, Reza Babanezhad Guo, Xingzhi Yang, Deqing Xiao, Yanghua The School of Data Science Fudan University China Shanghai Key Laboratory of Data Science School of Computer Science Fudan University China Department of Computer Science Stony Brook University United States Samsung SAIT AI Lab

Given the damping factor α and precision tolerance ϵ, Andersen et al. [2] introduced Approximate Personalized PageRank (APPR), the de facto local method for approximating the PPR vector, with runtime bounded by Θ(1/(αϵ)) independent of the graph size. Recently, Fountoulakis & Yang [12] asked whether faster local algorithms could be developed using Õ(1/(√αϵ)) operations. By noticing that APPR is a local variant of Gauss-Seidel, this paper explores the question of whether standard iterative solvers can be effectively localized. We propose to use the locally evolving set process, a novel framework to characterize the algorithm locality, and demonstrate that many standard solvers can be effectively localized. Let vol(St) and γt be the running average of volume and the residual ratio of active nodes St during the process. We show vol(St)/γt ≤ 1/ϵ and prove APPR admits a new runtime bound Õ(vol(St)/(αγt)) mirroring the actual performance. Furthermore, when the geometric mean of residual reduction is Θ(√α), then there exists c ∈ (0, 2) such that the local Chebyshev method has runtime Õ(vol(St)/(√α(2 − c))) without the monotonicity assumption. Numerical results confirm the efficiency of this novel framework and show up to a hundredfold speedup over corresponding standard solvers on real-world graphs. © 2024 Neural information processing systems foundation. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Learning from Highly Sparse Spatio-temporal data 38

Learning from Highly Sparse Spatio-temporal Data

引用

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Deng, Leyan Lian, Defu Wu, Chenwang Chen, Enhong School of Artificial Intelligence and Data Science University of Science and Technology of China China School of Computer Science and Technology University of Science and Technology of China China

Incomplete spatio-temporal data in the real world has spawned much research. However, existing methods often utilize iterative message-passing across temporal and spatial dimensions, resulting in substantial information loss and high computational cost. We provide a theoretical analysis revealing that such iterative models are susceptible to data and graph sparsity, causing unstable performances on different datasets. To overcome these limitations, we introduce a novel method named One-step Propagation and Confidence-based Refinement (OPCR). In the first stage, OPCR leverages inherent spatial and temporal relationships by employing a sparse attention mechanism. These modules propagate limited observations directly to the global context through one-step imputation, which is theoretically affected only by data sparsity. Following this, we assign confidence levels to the initial imputations by correlating missing data with valid data. This confidence-based propagation refines the separate spatial and temporal imputation results through spatio-temporal dependencies. We evaluate the proposed model across various downstream tasks involving highly sparse spatio-temporal data. Empirical results indicate that our model outperforms state-of-the-art imputation methods, demonstrating its effectiveness and robustness. © 2024 Neural information processing systems foundation. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

AVF-Lip: High-Fidelity Talking Face Generation via Audio-Visual Fusion 7

AVF-Lip: High-Fidelity Talking Face Generation via Audio-Vis...

引用

7th International Conference on Pattern Recognition and Artificial Intelligence, PRAI 2024

作者： Wu, Mengduo Zhao, Sirui Wu, Tong Xu, Yifan Xu, Tong Chen, Enhong University of Science and Technology of China School of Computer Science and Technology Hefei China University of Science and Technology of China School of Data Science Hefei China

ISBN: (纸本)9798350350890

Talking face generation aims to create realistic facial videos with lips precisely synchronized to the input audio, finding broad applications in human-computer interaction, virtual avatars, and video conferencing. Despite significant advancements in previous works, existing methods continue to face challenges in achieving accurate lip synchronization and high visual quality. To address these challenges, we propose a talking face generation framework named AVF-Lip that supports higher resolution and accurate lip synchronization. Specifically, an audio-visual fusion module (AVFM) is first customized to improve lip synchronization performance, and this is further enhanced by introducing a dynamic and fine-grained face mask. Subsequently, to address the problem of lip distortion in high-resolution facial images, we design a set of loss functions based on face sketches to enhance the generation of the lip areas. Extensive experiments demonstrate that our AVF-Lip outperforms state-of-the-art methods in lip synchronization and delivers competitive visual quality, positioning it as a promising solution for high-resolution and high-fidelity talking face generation. Project page: https://***/ © 2024 IEEE.

关键词： Video conferencing

来源：评论

学校读者我要写书评

暂无评论

CMDM: A Control Motion Diffusion Model for 2D Digital Human Motion Video Generation 7

CMDM: A Control Motion Diffusion Model for 2D Digital Human ...

引用

7th International Conference on Pattern Recognition and Artificial Intelligence, PRAI 2024

作者： Wu, Tong Zhao, Sirui Jin, Siyuan Xu, Tong Chen, Enhong School of Data Science University of Science and Technology of China Hefei China School of Computer Science and Technology University of Science and Technology of China Hefei China

ISBN: (纸本)9798350350890

In recent years, 2D digital human motion generation (DHMG) is becoming increasingly crucial for many areas such as virtual live broadcasting and film production. Although a lot of effort has been invested in DHMG, the complexity and diversity of human movements make it difficult to generate motions for arbitrary poses, and it is challenging to avoid distortion in the generated hand regions. To address these challenges, in this paper, we propose Control Motion Diffusion Model (CMDM), a diffusion based method for generating 2D digital human motion videos with arbitrary given poses. In CMDM, a novel Pose Control Module (PCM) is customized to fully encode pose and character information, generating multi-level and multi-resolution embedding, thereby ensuring that the model can learn more precise human motion features. In addition, data augmentation and VAE fine-tuning training strategies are adopted to decouple character and pose features, improving the model's ability to generate unseen poses. Finally, comprehensive experiments have been completed to demonstrate the effectiveness of the proposed method, and the proposed CMDM has achieved significant improvement in both subjective and objective evaluations. © 2024 IEEE.

关键词： Digital elevation model

来源：评论

学校读者我要写书评

暂无评论

Visualization and Simulation of Any MPU Architecture Design with Meta-Instruction 23

Visualization and Simulation of Any MPU Architecture Design ...

引用

23rd International Symposium on Communications and Information Technologies, ISCIT 2024

作者： Morimoto, Tomoyuki Wang, Ziming Luo, Jiahui Ogawa, Sota Tsutsumi, Toshiyuki Meiji University Department of Computer Science School of Science and Technology Kanagawa Japan Graduate School of Science and Technology Meiji University Computer Science Program Kanagawa Japan

ISBN: (纸本)9798350353952

Our conventional architecture design tool converted instructions of microprocessor (MPU) into meta-instructions with both semantic and functional expressions, and it displayed a circuit diagram of the meta-instruction. The tool was able to perform dynamic simulation only on an educational processor called COMET. However, the meta-instruction must be able to express the arbitrary instruction of at least different significant existing MPUs in order to design arbitrary MPUs. Furthermore, it should be possible to visualize all instructions and then execute them. Therefore, we have improved the meta-instruction so that the architecture design tool applies to major MPUs: MIPS, ARM, and RISC-V in addition to COMET. The improved architecture design tool visualizes and simulates the circuit diagram of the improved meta-instruction. The result shows the possibility of the meta-instruction-based architecture design tool being able to design any MPU. © 2024 IEEE.

关键词： Microprocessor chips

来源：评论

学校读者我要写书评

暂无评论

Secured Provisioning of Multimedia Contents in Cloud Platform Using Cryptographic Approaches 1

Secured Provisioning of Multimedia Contents in Cloud Platfor...

引用

1st International Conference on computer science and Emerging Technologies, CSET 2023

作者： Arul Mary, S.A. Sahaaya Nevgi, Akshaj Nair, Rhea Ranjith Sharukh Bhathena, Jehan Chaudhry, Angad Jessika, S. Sharon School of Computer Science and Engineering Vit India Computer Science and Engineering India Computer Science with Specialization in Data Science India

ISBN: (纸本)9798350341737

Cloud computing is a technology that allows the utilisation of a vast network of computers that are distributed and run in parallel with one another. The management of the multimedia files presents challenges and the significant of which is the prevalence of attackers and users with malicious intent. This article presents the conceptual foundation for a four-tiered data security paradigm that can be implemented in cloud computing. The proposed framework employs a DNA-computing-based password or secret key with a length of 1024 bits for the purpose of encrypting the user private multimedia file or correspondence. The DNABMS algorithm is applied in the cryptographic generation of digital identities as well as pseudorandom number generators. The results of the testing that was done on the model have validated both its reliability and its efficacy, demonstrating that the model that was developed is both reliable and effective. Experiments have demonstrated that the developed method is superior to other well-known existing strategies in terms of both effectiveness and efficiency. © 2023 IEEE.

关键词： Cryptography

来源：评论

学校读者我要写书评

暂无评论

NeuroProbe: A Performance Profiling Feed Forward Neural Network Simulator for Hardware Interaction Analysis

引用

SN computer science 2025年第5期6卷 1-22页

作者： Ayyappan, Binu Kumar, G. Santhosh Department of Computer Science Cochin University of Science and Technology Kochi India Department of Artificial Intelligence and Data Science Rajagiri School of Engineering & Technology Kochi India

NeuroProbe is a simple neural network simulator designed by authors specifically for educational purposes focusing on simulating inference phase on a computationally capable embedded hardware, aiming to provide a deeper understanding of hardware interactions and performance improvement. This simulator incorporates performance profiling capabilities, allowing students and educators to analyze and optimize the performance of a simulated trained neural network. By simulating the intricate hardware interactions, NeuroProbe enables users to gain insights into how various components impact the overall performance of neural networks. Using this simulator, learners can experiment with different configurations and architectures, identify potential bottlenecks, and explore optimization strategies to enhance neural network performance before deploying the model on an Edge Computing environment. It also enriches computer science education by enabling learners to gain deeper insights into hardware interactions and performance improvement. By identifying performance bottlenecks, learners can refine their understanding of neural networks and develop skills in optimizing their models for enhanced performance. In conclusion, NeuroProbe empowers users to bridge the gap between theoretical knowledge and practical implementation of neural network on an embedded device, fostering a more comprehensive understanding of neural networks and their hardware requirements. Currently, it is focused solely on single-threaded CPU-based computation enabling SSE, AVX. At present, NeuroProbe does not support SIMD extensions such as AVX-FMA, AVX-512, multi-threaded computation, and accelerators such as NPU, TPU, and GPU. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

关键词： Hardware–software interactions Neural network simulator NeuroProbe Performance bottlenecks Performance profiling Simulations

来源：评论

学校读者我要写书评

暂无评论

Scaling Up Multi-Agent Reinforcement Learning via Graph Decomposition Invariant Network 5

Scaling Up Multi-Agent Reinforcement Learning via Graph Deco...

引用

5th International Seminar on Artificial Intelligence, Networking and Information Technology, AINIT 2024

作者： Fu, Hongyi Ji, Jianmin School of Data Science University of Science and Technology of China Hefei China School of Computer Science and Technology University of Science and Technology of China Hefei China

ISBN: (纸本)9798350385557

This paper focuses on the dilemma faced by largescale multi-agent systems. With the increase of agent size, the policy search space grows exponentially, reaching the task complexity that is difficult for the current mainstream multiagent algorithms to deal with, so that the optimal strategy is converged in large-scale multi-agent systems. In order to make the learning process of large-scale multi-agent tasks more smooth, many researchers have introduced the method of course learning, setting the scale from small to large courses, compared with learning training from scratch, this way can continue to inherit the knowledge learned in the smaller scale tasks, making the training smoother. However, in order to realize the multiagent course learning from small to large scale, we face the following difficulties. First, the neural network structure we designed can adapt to the observation input of agents of different scales, rather than only applicable to a fixed scale. Second, in the multi-stage course learning, due to the network's inefficient representation of multi-agent systems, the knowledge learned in small-scale courses cannot be effectively utilized in large-scale courses, and even the problem of catastrophic forgetting occurs in the neural network. Specifically, we view multi-agent systems from a graph perspective and derive the Graph Decomposition Invariant (GDI) for the first time. We prove that neural network architectures satisfying the GDI are of great help in realizing efficient course learning. We design a model architecture called GDI model suitable for multi-agent curriculum learning, which satisfies the graph decomposition invariant proposed by us. Finally, we designed small-scale to large-scale multi-agent tasks on two experimental platforms, StarCraft and Neural MMO, and conducted a large number of comparative experiments, which verified that our proposed GDI model has a great promotion effect on course learning. © 2024 IEEE.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

AIGLLM: An Action Instruction Generation Method with Visually Enhanced LLM 7

AIGLLM: An Action Instruction Generation Method with Visuall...

引用

7th International Conference on Pattern Recognition and Artificial Intelligence, PRAI 2024

作者： Xia, Yudong Zhao, Sirui Wu, Tong Tang, Huaying Xu, Tong University of Science and Technology of China School of Data Science Hefei China University of Science and Technology of China School of Computer Science and Technology Hefei China

ISBN: (纸本)9798350350890

Action Instruction Generation (AIG) is a key step in semantic-driven action sequence generation, which aims to embed text instructions describing specific actions into the driving text. AIG is a challenging task that requires accurately understanding the potential action semantics contained in the driving text. Besides, the challenge is exacerbated by the lack of paired text-action instructions data. To address these challenges and motivated by the powerful capabilities of LLMs in text generation, we propose AIGLLM, a novel AIG framework with visually enhanced Large Language Model (LLM). Specifically, we perform low-rank adaptation (LoRA) on the off-the-shelf LLM to enable the model understand action semantics and generate action instructions at appropriate positions within the original text sequence. Particularly, to address the current lack of data that can be used to fine-tune LLMs for AIG, we construct a dataset containing 12,000 text-action instruction pairs and 15,000 image-action instruction pairs. Subsequently, a visual module is designed to enable the LLM to learn action knowledge from external visual signals of actions, thereby enhancing the model's understanding of action semantics. Finally, we conduct extensive experiments, where the experimental results and evaluations demonstrate that our method has an impressive key benefit: improved accurate action instruction embedding. © 2024 IEEE.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

The Implicit Bias of Adam on Separable data 38

The Implicit Bias of Adam on Separable Data

引用

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Zhang, Chenyang Zou, Difan Cao, Yuan Department of Statistics and Actuarial Science School of Computing and Data Science The University of Hong Kong Hong Kong Department of Computer Science School of Computing and Data Science Institute of Data Science The University of Hong Kong Hong Kong Department of Statistics and Actuarial Science School of Computing and Data Science Department of Mathematics The University of Hong Kong Hong Kong

Adam has become one of the most favored optimizers in deep learning problems. Despite its success in practice, numerous mysteries persist regarding its theoretical understanding. In this paper, we study the implicit bias of Adam in linear logistic regression. Specifically, we show that when the training data are linearly separable, the iterates of Adam converge towards a linear classifier that achieves the maximum ∞-margin in direction. Notably, for a general class of diminishing learning rates, this convergence occurs within polynomial time. Our result shed light on the difference between Adam and (stochastic) gradient descent from a theoretical perspective. © 2024 Neural information processing systems foundation. All rights reserved.

关键词：

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：