检索结果-内蒙古大学图书馆

2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024

作者： Wang, Pengyu Zhang, Dong Li, Linyang Tan, Chenkun Wang, Xinghao Zhang, Mozhi Ren, Ke Jiang, Botian Qiu, Xipeng School of Computer Science Fudan University Shanghai Key Laboratory of Intelligent Information Processing Fudan University

ISBN: (纸本)9798891761643

As large language models (LLMs) rapidly evolve, they are increasingly being customized through fine-tuning to suit the specific needs of various applications. A critical aspect of this advancement is the alignment process, which ensures that these models perform tasks in ways that align with human values and expectations. Current alignment methods, such as direct preference optimization (DPO) and reinforcement learning from human feedback (RLHF), focus primarily on alignment during training phase. However, these methods often involve complex and resource-intensive training processes, posing significant challenge for their implementation. Therefore, we propose InferAligner, a simple yet effective method for harmlessness alignment during inference phase. InferAligner decouples harmlessness from helpfulness. During the training phase, it focuses solely on enhancing the target model's capabilities on downstream tasks. In the inference phase, it utilizes safety steering vectors extracted from the aligned model to guide the target model towards harmlessness alignment. Experimental results show that our method can be very effectively applied to domain-specific models in finance, medicine, and mathematics, as well as to multimodal large language models (MLLMs) such as LLaVA. It significantly diminishes the attack success rate (ASR) of both harmful instructions and jailbreak instructions, while maintaining almost unchanged performance in downstream tasks. © 2024 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

SPEECHTOKENIZER: UNIFIED SPEECH TOKENIZER FOR SPEECH LANGUAGE MODELS 12

SPEECHTOKENIZER: UNIFIED SPEECH TOKENIZER FOR SPEECH LANGUAG...

引用

12th International Conference on Learning Representations, ICLR 2024

作者： Zhang, Xin Zhang, Dong Li, Shimin Zhou, Yaqian Qiu, Xipeng School of Computer Science Fudan University Shanghai Key Laboratory of Intelligent Information Processing Fudan University

Current speech large language models build upon discrete speech representations, which can be categorized into semantic tokens and acoustic tokens. However, existing speech tokens are not specifically designed for speech language modeling. To assess the suitability of speech tokens for building speech language models, we established the first benchmark, SLMTokBench. Our results indicate that neither semantic nor acoustic tokens are ideal for this purpose. Therefore, we propose SpeechTokenizer, a unified speech tokenizer for speech large language models. SpeechTokenizer adopts the Encoder-Decoder architecture with residual vector quantization (RVQ). Unifying semantic and acoustic tokens, SpeechTokenizer disentangles different aspects of speech information hierarchically across different RVQ layers. Furthermore, We construct a Unified Speech Language Model (USLM) leveraging SpeechTokenizer. Experiments show that SpeechTokenizer performs comparably to EnCodec in speech reconstruction and demonstrates strong performance on the SLMTokBench benchmark. Also, USLM outperforms VALL-E in zero-shot Text-to-Speech tasks. Code and models are available at https://***/ZhangXInFD/SpeechTokenizer/. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Understanding News Background Information Seeking from Readers' Perspective 25

Understanding News Background Information Seeking from Reade...

引用

25th IEEE International Conference on computer Supported Cooperative Work in Design, CSCWD 2022

作者： Feng, Zhenni Ding, Xianghua Jiang, Yanqi Wu, Yiying Zhang, Peng Fudan University School of Computer Science Shanghai Key Laboratory of Data Science Shanghai China University of Glasgow School of Computer Science United Kingdom

ISBN: (纸本)9781665405270

Background information is very important for news reading and understanding, especially with today's rapid new media development for news production and consumption. When researchers actively develop tools to automatically provide back-ground information for people, we still have little understanding of how people, from readers' perspective, engage with news and what background information is needed. To address such a gap, we conducted a study with a news reading platform, observed participants' news reading behaviors using the platform and interviewed them over the process. Our study highlighted that news reading is a learning activity between leisure and study, and their background information searching behavior is part of Gestalt knowing - to make their existing knowledge more complete, by deepening their prior understandings, broadening their horizons, filling the gaps, clearing their confusion, as well as tracking the progress of an event as a whole, so a more holistic understanding could be achieved. We end by discussing implications of our findings. © 2022 IEEE.

关键词： Leisure

来源：评论

学校读者我要写书评

暂无评论

Efficient ILT via Multigrid-Schwartz Method 24

Efficient ILT via Multigrid-Schwartz Method

引用

61st ACM/IEEE Design Automation Conference, DAC 2024

作者： Sun, Shuyuan Yang, Fan Yu, Bei Shang, Li Zhou, Dian Zeng, Xuan State Key Laboratory of Integrated Chips and Systems Fudan University Shanghai China School of Microelectronics Fudan University Shanghai China China and Shanghai Key Laboratory of Data Science School of Computer Science Fudan University Shanghai China Chinese University of Hong Kong Hong Kong University of Texas DallasTX United States

ISBN: (纸本)9798400706011

Inverse Lithography Technology (ILT) is an important Resolution Enhancement Technology (RET) in chip manufacturing. Due to the high computational demands of ILT, large-scale layouts are typically partitioned into smaller tiles for independent processing. In this paper, we propose a multigrid-Schwarz method to overcome challenges in tile assembly. Experimental results show that our approach achieves comparable performance to the full-chip ILT, offering increased parallelizability and speedup in parallel mode. Unlike the traditional divide-and-conquer algorithm, it effectively alleviates discontinuities of tile stitching, preventing manufacturing failures. © 2024 Copyright is held by the owner/author(s). Publication rights licensed to ACM.

关键词： Inverse problems

来源：评论

学校读者我要写书评

暂无评论

Research on Improved MobileViT Image Tamper Localization Model

引用

computers, Materials & Continua 2024年第8期80卷 3173-3192页

作者： Jingtao Sun Fengling Zhang Huanqi Liu Wenyan Hou School of Computer Science and Technology Xi’an University of Posts and TelecommunicationsXi’an710121China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Xi’an University of Posts and TelecommunicationsXi’an710121China

As image manipulation technology advances rapidly,the malicious use of image tampering has alarmingly escalated,posing a significant threat to social *** the realm of image tampering localization,accurately localizing limited samples,multiple types,and various sizes of regions remains a multitude of *** issues impede the model’s universality and generalization capability and detrimentally affect its *** tackle these issues,we propose FL-MobileViT-an improved MobileViT model devised for image tampering *** proposed model utilizes a dual-stream architecture that independently processes the RGB and noise domain,and captures richer traces of tampering through dual-stream ***,the model incorporating the Focused Linear Attention mechanism within the lightweight network(MobileViT).This substitution significantly diminishes computational complexity and resolves homogeneity problems associated with traditional Transformer attention mechanisms,enhancing feature extraction diversity and improving the model’s localization *** comprehensively fuse the generated results from both feature extractors,we introduce the ASPP architecture for multi-scale feature *** facilitates a more precise localization of tampered regions of various ***,to bolster the model’s generalization ability,we adopt a contrastive learning method and devise a joint optimization training strategy that leverages fused features and captures the disparities in feature distribution in tampered *** strategy enables the learning of contrastive loss at various stages of the feature extractor and employs it as an additional constraint condition in conjunction with cross-entropy *** a result,overfitting issues are effectively alleviated,and the differentiation between tampered and untampered regions is *** evaluations on five benchmark datasets(IMD-20,CASIA,NIST-16,Columbia and Coverage)validat

关键词： Image tampering localization focused linear attention mechanism MobileViT contrastive loss

来源：评论

学校读者我要写书评

暂无评论

Label distribution similarity-based noise correction for crowdsourcing

引用

Frontiers of computer science 2024年第5期18卷 71-82页

作者： Lijuan REN Liangxiao JIANG Wenjun ZHANG Chaoqun LI School of Computer Science China University of GeosciencesWuhan 430074China Key Laboratory of Artificial Intelligence Ministry of EducationShanghai 200240China School of Mathematics and Physics China University of GeosciencesWuhan 430074China

In crowdsourcing scenarios,we can obtain each instance's multiple noisy labels from different crowd workers and then infer its integrated label via label *** spite of the effectiveness of label aggregation methods,there still remains a certain level of noise in the integrated ***,some noise correction methods have been proposed to reduce the impact of noise in recent ***,to the best of our knowledge,existing methods rarely consider an instance's information from both its features and multiple noisy labels simultaneously when identifying a noise *** this study,we argue that the more distinguishable an instance's features but the noisier its multiple noisy labels,the more likely it is a noise *** on this premise,we propose a label distribution similarity-based noisecorrection(LDSNC)*** measure whether an instance's features are distinguishable,we obtain each instance's predicted label distribution by building multiple classifiers using instances'features and their integrated *** measure whether an instance's multiple noisy labels are noisy,we obtain each instance's multiple noisy label distribution using its multiple noisy ***,we use the Kullback-Leibler(KL)divergence to calculate the similarity between the predicted label distribution and multiple noisy label distribution and define the instance with the lower similarity as a noise *** extensive experimental results on 34 simulated and four real-world crowdsourced datasets validate the effectiveness of our method.

关键词： crowdsourcing learning noise correction label distribution similarity kullback-leibler divergence

来源：评论

学校读者我要写书评

暂无评论

An Efficient and Fast Area Optimization Approach for Mixed Polarity Reed-Muller Logic Circuits

引用

Chinese Journal of Electronics 2024年第5期33卷 1165-1180页

作者： Yuhao ZHOU Zhenxue HE Jianhui JIANG Xiaojun ZHAO Fan ZHANG Limin XIAO Xiang WANG School of Software Engineering Tongji University Key Laboratory of Agricultural Big Data of Hebei Province Hebei Agricultural University School of Computer Science and Engineering Beihang University School of Electronic and Information Engineering Beihang University

Area has become one of the main bottlenecks restricting the development of integrated circuits. The area optimization approaches of existing XNOR/OR-based mixed polarity Reed-Muller(MPRM) circuits have poor optimization effect and efficiency. Given that the area optimization of MPRM logic circuits is a combinatorial optimization problem, we propose a whole annealing adaptive bacterial foraging algorithm(WAA-BFA), which includes individual evolution based on Markov chain and Metropolis acceptance criteria, and individual mutation based on adaptive probability. To address the issue of low conversion efficiency in existing polarity conversion approaches, we introduce a fast polarity conversion algorithm(FPCA). Moreover, we present an MPRM circuits area optimization approach that uses the FPCA and WAA-BFA to search for the best polarity corresponding to the minimum circuits area. Experimental results demonstrate that the proposed MPRM circuits area optimization approach is effective and can be used as a promising EDA tool.

关键词： Integrated circuits Microorganisms Annealing Logic circuits Circuits Benchmark testing Delays

来源：评论

学校读者我要写书评

暂无评论

Leaving None Behind: data-Free Domain Incremental Learning for Major Depressive Disorder Detection

引用

IEEE Transactions on Affective Computing 2024年第2期16卷 758-770页

作者： Chen, Tao Guo, Yanrong Hao, Shijie Hong, Richang Hefei University of Technology Key Laboratory of Knowledge Engineering with Big Data China Hefei University of Technology Ministry of Education and School of Computer Science and Information Engineering Hefei230009 China

While deep learning techniques have shown promising performance in the Major Depressive Disorder (MDD) detection task, they still face limitations in real-world scenarios. Specifically, given the data scarcity, some efforts have resorted to aggregating data from different domains to expand the data volume. However, their effectiveness is currently limited by the domain gap and data privacy. Additionally, the class imbalance issue is particularly severe in our application, leading to biased classifying performance accordingly. To address these challenges, we propose data-Free Domain Incremental Learning for the MDD detection (DIL-MDD) task, accommodating multiple feature distributions by only accessing well-trained models from previous domains and the data in the current domain. Specifically, DIL-MDD consists of two key modules: Adaptive Class-tailored Threshold Learning (ACTL) and data-Free Domain Alignment (DFDA). The first module measures the discrepancy between the outputs of two sequential domains, based on which we learn a class-tailored threshold adaptively. Building on this, we differentiate between samples that either exhibit similarities or dissimilarities with the previous domain, where this similar sample set is identified to investigate the feature distribution of the historical data. The second module imposes an alignment constraint to narrow the gap between these two sample sets, thereby exploring the expertise of the previous domain. To validate the effectiveness of the proposed method, we conduct extensive experiments on the public MDD datasets, i.e., DAIC-WOZ, MODMA, and CMDC. We also apply our method to another mental health condition, Autism Spectrum Disorder (ASD), to further demonstrate its applicability. Finally, the ablation studies validate the superiority of the proposed modules. © 2010-2012 IEEE.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

On construction of strong orthogonal arrays and column-orthogonal strong orthogonal arrays of strength two plus

引用

science China Mathematics 2025年第5期68卷 1219-1242页

作者： Bochuan Jiang Yaping Wang Fasheng Sun School of Mathematics and Statistics Beijing Jiaotong UniversityBeijing100044China Key Laboratory of Advanced Theory and Application in Statistics and Data Science Ministry of EducationSchool of StatisticsEast China Normal UniversityShanghai200062China Key Laboratory for Applied Statistics of Ministry of Education and School of Mathematics and Statistics Northeast Normal UniversityChangchun130024China

computer experiments require space-filling designs with good low-dimensional projection *** orthogonal arrays are a type of space-filling design that provides better stratifications in low dimensions than ordinary orthogonal *** this paper,we address the problem of constructing strong orthogonal arrays and column-orthogonal strong orthogonal arrays of strength two *** methods typically rely on regular designs or specific nonregular designs as base orthogonal arrays,limiting the sizes of the final ***,we propose two general methods that are easy to implement and applicable to a wide range of base orthogonal *** methods produce space-filling designs that can accommodate a large number of factors,provide significant flexibility in terms of run sizes,and possess appealing low-dimensional projection ***,these designs are ideal for computer experiments.

关键词： computer experiment difference scheme strong orthogonal array column-orthogonality space-filling property

来源：评论

学校读者我要写书评

暂无评论

ControlVideo: conditional control for one-shot text-driven video editing and beyond

引用

science China(Information sciences) 2025年第3期68卷 150-162页

作者： Min ZHAO Rongzhen WANG Fan BAO Chongxuan LI Jun ZHU Department of Computer Science and Technology Institute for AI Tsinghua-Bosch Joint ML CenterTsinghua Laboratory of Brain and Intelligence Lab Tsinghua University ShengShu Technology Gaoling School of Artificial Intelligence Renmin University of China Beijing Key Laboratory of Big Data Management and Analysis Methods Pazhou Laboratory (Huangpu)

This paper presents ControlVideo for text-driven video editing — generating a video that aligns with a given text while preserving the structure of the source video. Building on a pre-trained text-to-image diffusion model, ControlVideo enhances the fidelity and temporal consistency by incorporating additional conditions(such as edge maps), and fine-tuning the key-frame and temporal attention on the source video-text pair via an in-depth exploration of the design space. Extensive experimental results demonstrate that ControlVideo outperforms various competitive baselines by delivering videos that exhibit high fidelity w.r.t. the source content, and temporal consistency, all while aligning with the text. By incorporating low-rank adaptation layers into the model before training, ControlVideo is further empowered to generate videos that align seamlessly with reference images. More importantly, ControlVideo can be readily extended to the more challenging task of long video editing(e.g., with hundreds of frames), where maintaining long-range temporal consistency is crucial. To achieve this, we propose to construct a fused ControlVideo by applying basic ControlVideo to overlapping short video segments and key frame videos and then merging them by pre-defined weight functions. Empirical results validate its capability to create videos across 140 frames, which is approximately 5.83 to 17.5 times more than what previous studies achieved. The code is available at https://***/thu-ml/controlvideo.

关键词： diffusion models controllable generation text-driven editing video editing long video editing

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：