检索结果-内蒙古大学图书馆

Dynamic Expansion Diffusion Learning for Lifelong Generative Modelling 39

学校读者我要写书评

暂无评论

Dynamic Expansion Diffusion Learning for Lifelong Generative...

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Ye, Fei Bors, Adrian G. Zhang, Kun School of Information and Software Engineering University of Electronic Science and Technology of China Chengdu China Department of Computer Science University of York YorkYO10 5GH United Kingdom MBZUAI Abu Dhabi United Arab Emirates Carnegie Mellon University PittsburghPA United States

ISBN: (纸本)157735897X

The diffusion model has lately been shown to achieve remarkable performances through its ability of generating high quality images. However, current diffusion model studies consider only learning from a single data distribution, resulting in catastrophic forgetting when attempting to learn new data. In this paper, we explore a more realistic learning scenario where training data is continuously acquired. We propose the Dynamic Expansion Diffusion Model (DEDM) for addressing catastrophic forgetting and data distribution shifts under Online Task-Free Continual Learning (OTFCL) paradigm. New diffusion components are added to a mixture model following the evaluation of a criterion which compares the probabilistic representation of the new data with the existing knowledge of the DEDM model. In addition, to maintain an optimal architecture, we propose a component discovery approach that ensures the diversity of knowledge while minimizing the total number of parameters in the DEDM. Furthermore, we show how the proposed DEDM can be implemented as a teacher module in a unified framework for representation learning. In this approach, knowledge distillation is proposed for training a student module aiming to compress the teacher’s knowledge into the latent space of the student. Copyright © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词：

AdaVITS: Tiny VITS for Low Computing Resource Speaker Adaptation

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Song, Kun Xue, Heyang Wang, Xinsheng Cong, Jian Zhang, Yongmao Xie, Lei Yang, Bing Zhang, Xiong Su, Dan School of Computer Science School of Software Northwestern Polytechnical University Xi’an China Cloud and Smart Industries Group Tencent Technology Co. Ltd. China

Speaker adaptation in text-to-speech synthesis (TTS) is to finetune a pre-trained TTS model to adapt to new target speakers with limited data. While much effort has been conducted towards this task, seldom work has been performed for low computational resource scenarios due to the challenges raised by the requirement of the lightweight model and less computational complexity. In this paper, a tiny VITS-based [1] TTS model, named AdaVITS, for low computing resource speaker adaptation is proposed. To effectively reduce the parameters and computational complexity of VITS, an inverse short-time Fourier transform (iSTFT)-based wave construction decoder is proposed to replace the upsampling-based decoder which is resource-consuming in the original VITS. Besides, NanoFlow is introduced to share the density estimate across flow blocks to reduce the parameters of the prior encoder. Furthermore, to reduce the computational complexity of the textual encoder, scaled-dot attention is replaced with linear attention. To deal with the instability caused by the simplified model, we use phonetic posteriorgram (PPG) as a frame-level linguistic feature for supervising the model process from phoneme to spectrum. Experiments show that AdaVITS can generate stable and natural speech in speaker adaptation with 8.97M model parameters and 0.72 GFlops computational complexity. 1 Copyright © 2022, The Authors. All rights reserved.

关键词： Speech synthesis

PCA: Semi-Supervised Segmentation with Patch Confidence Adversarial Training

学校读者我要写书评

暂无评论

IEEE Transactions on Network science and Engineering 2025年

作者： Xu, Zhenghua Yang, Runhe Xu, Zihang Zhang, Shuo Yang, Yuchen Liu, Weipeng Xu, Weichao Chen, Junyang Lukasiewicz, Thomas Leung, Victor C. M. Hebei University of Technology State Key Laboratory of Reliability and Intelligence of Electrical Equipment School of Health Sciences and Biomedical Engineering Tianjin China The University of Hong Kong Department of Electrical and Electronic Engineering Hong Kong The Johns Hopkins University Department of Applied Mathematics and Statistic Baltimore United States Hebei University of Technology School of Artificial Intelligence Tianjin China Hebei Provincial Hospital of Traditional Chinese Medicine Digestive Disease Center Shijiazhuang China Shenzhen University College of Computer Science and Software Engineering Shenzhen China University of Oxford Department of Computer Science University of Oxford United Kingdom Department of Computer Science United Kingdom Vienna University of Technology institute of Logic and Computation nstitute of Logic and Computation Vienna Austria

Deep-learning-based semi-supervised learning (SSL) methods have achieved a strong performance in medical image segmentation, which can alleviate doctors' expensive annotation by utilizing a large amount of unlabeled data. Unlike most existing semi-supervised learning methods, adversarial training methods distinguish samples from different sources by learning the data distribution of the segmentation map, leading the segmenter to generate more accurate predictions. We argue that the current performance restrictions for such approaches are the problems of feature extraction and learning preferences. In this paper, we propose a new semi-supervised adversarial method called Patch Confidence Adversarial Training (PCA) for medical image segmentation. The PCA method's discriminator penalizes patch-level structures, guiding the generator to optimize different patch areas, by leveraging pixel context, the generator is driven to focus on high-frequency features, making it harder to deceive the discriminator and easy to converge to an ideal state, which more effectively guides the segmenter to generate high-quality pseudo-labels. Furthermore, at the discriminator's input, we supplement image information constraints, making it simpler to fit the expected data distribution. Extensive experiments on the Automated Cardiac Diagnosis Challenge (ACDC) 2017 dataset and the Brain Tumor Segmentation (BraTS) 2019 challenge dataset show that our method outperforms the state-of-the-art semi-supervised methods, which demonstrates its effectiveness for medical image segmentation. © 2025 IEEE. All rights reserved.

关键词： Adversarial machine learning

Distributed Quantum Simulation

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Feng, Tianfeng Xu, Jue Yu, Wenjun Ye, Zekun Yao, Penghui Zhao, Qi QICI Quantum Information and Computation Initiative Department of Computer Science School of Computing and Data Science The University of Hong Kong Pokfulam Road Hong Kong State Key Laboratory for Novel Software Technology Nanjing University Hefei National Laboratory Nanjing 210023 China Hefei230088 China

Quantum simulation is a promising pathway toward practical quantum advantage by simulating large-scale quantum systems. In this work, we propose communication-efficient distributed quantum simulation protocols by exploring three quantum simulation algorithms, including the product formula, the truncated Taylor series, and the processing of quantum signals over a quantum network. Our protocols are further shown to be optimal by deriving a lower bound on the quantum communication complexity for distributed quantum simulations with respect to evolution time and the number of distributed quantum processing units. Additionally, our distributed techniques go beyond quantum simulation and are applied to distributed versions of Grover’s algorithms and quantum phase estimation. Our work not only paves the way for achieving a practical quantum advantage by scalable quantum simulation but also enlightens the design of more general distributed architectures across various physical systems for quantum computation. Copyright © 2024, The Authors. All rights reserved.

关键词： Quantum computers

PoW and PoS and Related Applications

学校读者我要写书评

暂无评论

PoW and PoS and Related Applications

2021 International Conference on Electronic Information Engineering and computer science, EIECS 2021

作者： Ouyang, Zhiqiang Shao, Jie Zeng, Yifeng School of Computer Science and Engineering Sun Yat-sen University Guangzhou China University of Washington Department of Mathematics WA United States Shool of Software Dalian University of Technology Dalian China

ISBN: (纸本)9781665416740

Consensus protocols can help guarantee security and sustainability to maintain the decentralized Blockchain system used in different scenarios and applications due to their different features. There are two main prevalent protocols, Proof of Work (PoW) used in Bitcoin and Proof of Stake (PoS) used in Ethereum. This paper reviews these two consensus protocols, PoW and PoS, and their related improvements and applications. We first introduce the basic concepts, implementations, advantages, and disadvantages of PoW and PoW and make a detailed comparison between these two protocols. We find that though providing security when honest participants are the majority, PoW may lead to excessive power consumption. While PoS is relatively insecure and leads to monopoly (centralization), though energy efficient by discarding complex hash operations. Therefore, some newly proposed consensus protocols have attempted to find a pleasant compromise between these two classic protocols. After a detailed analysis, we investigate the researches related to the defect improvements of PoS and PoW. Finally, we describe various applications of PoS and PoW, including cryptocurrencies and information storage. © 2021 IEEE.

关键词： Blockchain

Improving Deep Assertion Generation via Fine-Tuning Retrieval-Augmented Pre-trained Language Models

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Zhang, Quanjun Fang, Chunrong Zheng, Yi Zhang, Yaxin Zhao, Yuan Huang, Rubing Zhou, Jianyi Yang, Yun Zheng, Tao Chen, Zhenyu Department of Computing Technologies Swinburne University of Technology Melbourne Australia State Key Laboratory for Novel Software Technology Nanjing University Jiangsu Nanjing China School of Computer Science and Engineering Macau University of Science and Technology Jiangsu Nanjing China Huawei Cloud Computing Technologies Co. Ltd. Beijing China Shenzhen Research Institute of Nanjing University Guangdong Shenzhen China

Unit testing validates the correctness of the units of the software system under test and serves as the cornerstone in improving software quality and reliability. To reduce manual efforts in writing unit tests, some techniques have been proposed to generate test assertions automatically, including deep learning (DL)-based, retrieval-based, and integration-based ones. Among them, recent integration-based approaches inherit from both DL-based and retrieval-based approaches and are considered state-of-the-art. Despite being promising, such integration-based approaches suffer from inherent limitations, such as retrieving assertions with lexical matching while ignoring meaningful code semantics, and generating assertions with a limited training corpus. In this paper, we propose a novel Retrieval-Augmented Deep Assertion Generation approach, namely RetriGen, based on a hybrid assertion retriever and a pre-trained language model (PLM)-based assertion generator. Given a focal-test, RetriGen first builds a hybrid assertion retriever to search for the most relevant test-assert pair from external codebases. The retrieval process takes both lexical similarity and semantical similarity into account via a token-based and an embedding-based retriever, respectively. RetriGen then treats assertion generation as a sequence-to-sequence task and designs a PLM-based assertion generator to predict a correct assertion with historical test-assert pairs and the retrieved external assertion. Although our concept is general and can be adapted to various off-the-shelf encoder-decoder PLMs, we implement RetriGen to facilitate assertion generation based on the recent CodeT5 model. We conduct extensive experiments to evaluate RetriGen against six state-of-the-art approaches across two large-scale datasets and two metrics. The experimental results demonstrate that RetriGen achieves 57.66% and 73.24% in terms of accuracy and CodeBLEU, outperforming all baselines with an average improvement of 50.66

关键词： Semantics

ESENet: A Human Behavior Recognition Model Based on Extended Squeeze-and-Excitation Network 33rd

学校读者我要写书评

暂无评论

ESENet: A Human Behavior Recognition Model Based on Extended...

33rd International Conference on computer Animation and Social Agents, CASA 2020

作者： Chen, Lei Liu, Rui Zhou, Dongsheng Yang, Xin Zhang, Qiang Wei, Xiaopeng School of Software Engineering Dalian University Dalian China School of Computer Science and Technology Dalian University of Technology Dalian China

ISBN: (纸本)9783030634254

Human behavior recognition in video sequences is an important research problem in computer vision, and has attracted a lot of researchers’ attention in recent years. Because of its powerful ability in feature channels recalibration, the Squeeze-and-Excitation network (SE network) has been widely used in video behavior recognition models. However, most of these methods use the SE network directly to recalibrate different feature channels of a single frame image, but ignore the temporal dimension of videos. To address this problem, a new behavior recognition model based on extended Squeeze-and-Excitation network (named as ESENet) is proposed in this study. In order to improve the attention capability, the integration strategy of the 3DSE is also discussed in this study. Experiment results show that the proposed method in this paper achieves competitive performance on the challenging HMDB51, UCF101, and Something-Something v1 datasets. © 2020, Springer Nature Switzerland AG.

关键词： Behavioral research

TPE for JPEG images with Dynamic M-ary Decomposition and Adaptive Threshold Constraints

学校读者我要写书评

暂无评论

IEEE Transactions on Circuits and Systems for Video technology 2025年

作者： Ma, Yakun Chai, Xiuli Long, Guoqiang Gan, Zhihua Zhang, Yushu Henan University School of Artificial Intelligence Henan Engineering Research Center for Industrial Internet of Things Zhengzhou450046 China Henan University School of Software Intelligent Data Processing Engineering Research Center of Henan Province Institute of Intelligent Network System Kaifeng475004 China Nanjing University of Aeronautics and Astronautics College of Computer Science and Technology Nanjing211106 China

Traditional JPEG image encryption that prioritizes solely confidentiality fails to account for the pressing usability requirements of cloud-based environments, thus boosting the boom in thumbnail-preserving encryption (TPE) to balance image privacy and usability. However, existing TPE schemes for JPEG images face numerous challenges, such as insufficient security, inability to achieve lossless decryption, and high file extension. To address these challenges, we propose a TPE scheme based on dynamic M-ary decomposition and adaptive threshold constraints (TPE-MDTC). First, the valid ranges of quantized DC coefficients for JPEG images are determined. Then, a sum-preserving encryption method for quantized DC coefficients with compliance threshold constraints is designed using the bit-plane permutation to preserve thumbnails with high accuracy. Next, the introduction of dynamic M-ary decomposition effectively changes bit statistical characteristics preserved by bit-plane permutation, enhancing the ciphertext security. Finally, a quantized AC encryption method with RV (Run/Value) pair global permutation is proposed, effectively modifying the unit block features, thereby significantly improving the security and attack resistance of encrypted images. Experimental results show that the proposed TPE-MDTC scheme can reconstruct the original JPEG images without loss, and the generated ciphertext images exhibit significant advantages over previous schemes regarding file extension and security. © 1991-2012 IEEE.

关键词： Ciphertext

Commenting Higher-level Code Unit: Full Code, Reduced Code, or Hierarchical Code Summarization

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Sun, Weisong Zhang, Yiran Zhu, Jie Wang, Zhihui Fang, Chunrong Zhang, Yonglong Feng, Yebo Huang, Jiangping Wang, Xingya Jin, Zhi Liu, Yang Nanyang Technological University Singapore Yangzhou University China State Key Laboratory for Novel Software Technology Nanjing University China Chongqing University of Posts and Telecommunications China Nanjing Tech University China Ministry of Education School of Computer Science Peking University China

Commenting code is a crucial activity in software development, as it aids in facilitating future maintenance and updates. To enhance the efficiency of writing comments and reduce developers’ workload, the software engineering community has proposed various automated code summarization (ACS) techniques to automatically generate comments (also called summaries) for given code units. However, these ACS techniques, including the recent ACS techniques based on large language models (LLMs), primarily focus on generating summaries for code units at the method level. There is a significant lack of research on summarizing higher-level code units, such as file-level and module-level code units, despite the fact that summaries of these higher-level code units are highly useful for quickly gaining a macro-level understanding of software components, modules, and architecture. To fill this gap, in this paper, we conduct a systematic study on how to use LLMs for commenting higher-level code units, including file level and module level. These higher-level units are significantly larger than method-level ones, which poses challenges in handling long code inputs within LLM constraints and maintaining efficiency. To address these issues, we explore various summarization strategies for ACS of file-level and module-level code units, which can be categorized into three types: full code summarization, reduced code summarization, and hierarchical code summarization. As the name suggests, full code summarization involves inputting all the code of the higher-level code unit into LLMs as context. Reduced code summarization only inputs the key information, excluding details like method bodies. Hierarchical code summarization involves first generating summaries for lower-level code units (e.g., methods for files, files for modules), and then using those summaries to generate summaries for higher-level code units. We investigate three LLMs (including one general-purpose LLM and two specialized c

关键词： software design