检索结果-内蒙古大学图书馆

Adaptive Loose Optimization for Robust Question Answering

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Ma, Jie Wang, Pinghui Wang, Zewei Kong, Dechen Hu, Min Han, Ting Liu, Jun The Ministry of Education of Key Laboratory for Intelligent Networks and Network Security School of Cyber Science and Engineering Xi’an Jiaotong University Shaanxi Xi’an710049 China The Ministry of Education of Key Laboratory for Intelligent Networks and Network Security School of Automation Science and Engineering Xi’an Jiaotong University Shaanxi Xi’an710049 China The China Mobile Research Institute China The Shannxi Provincial Key Laboratory of Big Data Knowledge Engineering School of Computer Science and Technology Xi’an Jiaotong University Shaanxi Xi’an710049 China

Question answering methods are well-known for leveraging data bias, such as the language prior in visual question answering and the position bias in machine reading comprehension (extractive question answering). Current debiasing methods often come at the cost of significant in-distribution performance to achieve favorable out-of-distribution generalization ability, while non-debiasing methods sacrifice a considerable amount of out-of-distribution performance in order to obtain high in-distribution performance. Therefore, it is challenging for them to deal with the complicated changing real-world situations. In this paper, we propose a simple yet effective novel loss function with adaptive loose optimization, which seeks to make the best of both worlds for question answering. Our main technical contribution is to reduce the loss adaptively according to the ratio between the previous and the current optimization state on mini-batch training data. This loose optimization can be used to prevent non-debiasing methods from overlearning data bias while enabling debiasing methods to maintain slight bias learning. Experiments on the visual question answering datasets, including VQA v2, VQA-CP v1, VQA-CP v2, GQA-OOD, and the extractive question answering dataset SQuAD demonstrate that our approach enables QA methods to obtain state-of-the-art in- and out-of-distribution performance in most cases. The source code has been released publicly in https://***/reml-group/ALO. Copyright © 2023, The Authors. All rights reserved.

关键词： Machine learning

MDPL-net: Multi-layer Dictionary Learning Network with Added Skip Dense Connections

学校读者我要写书评

暂无评论

MDPL-net: Multi-layer Dictionary Learning Network with Added...

IEEE International Conference on data Mining (ICDM)

作者： Zhao Zhang Yulin Sun Zheng Zhang Yang Wang Lin Wu Meng Wang School of Computer Science and Technology Soochow University Suzhou China Ministry of Education Key Laboratory of Knowledge Engineering with Big Data School of Computer Science and Information Engineering Hefei University of Technology Hefei China Bio-Computing Research Center Harbin Institute of Technology (Shenzhen) Shenzhen China

ISBN: (数字)9781728183169

ISBN: (纸本)9781728183176

Dictionary learning (DL) is powerful for representation learning, while it fails to capture the deep hierarchical information hidden in data. In this paper, we propose a new generalized end-to-end mulita-layer representation learning architecture referred to as Multi-layer Dictionary Pair Learning Network (MDPL-net) for the deep sparse and hierarchical representation of images. To enable MDPL-net to conduct accurate classification, MDPL-net clearly integrates the skip connection end-to-end network and multi-layer deep sparse dictionary learning into a unified architecture. The representation learning module has several hidden DL blocks, where each hidden DL block has a dictionary pair learning (DPL) layer, a batch-norm layer and an activation function layer, and the DL blocks are connected in a feed-forward manner. To further improve the information flow and maintain the privileged features between different DL blocks, a novel skip dense connectivity pattern is deployed between hidden DL blocks, which can obtain more stable and discriminative features. The DPL layer jointly formulates the discriminative synthesis dictionary and analysis dictionary by minimizing reconstruction error within each batch over the feature maps from front layers. Extensive results on benchmark databases demonstrate the effectiveness of MDPL-net for discriminative representation and robust image classification.

关键词： Dictionaries databases Conferences Machine learning data mining Image reconstruction Image classification

ConsistencyDet: A Robust Object Detector with a Denoising Paradigm of Consistency Model

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Jiang, Lifan Wang, Zhihui Wang, Changmiao Li, Ming Leng, Jiaxu Wu, Xindong the College of Computer Science and Engineering Shandong University of Science and Technology Qingdao266510 China Shenzhen Research Institute of Big Data Shenzhen518172 China Key Laboratory of Intelligent Education Technology and Application Zhejiang Normal University Jinhua321004 China the Key Laboratory of Image Cognition Chongqing University of Posts and Telecommunications Chongqing400065 China Jiangsu Key Laboratory of Image and Video Understanding for Social Safety Nanjing University of Science and Technology Nanjing210094 China the Key Laboratory of Knowledge Engineering with Big Data the Ministry of Education of China China School of Computer Science and Information Technology Hefei University of Technology Hefei230009 China

Object detection, a quintessential task in the realm of perceptual computing, can be tackled using a generative methodology. In the present study, we introduce a novel framework designed to articulate object detection as a denoising diffusion process, which operates on the perturbed bounding boxes of annotated entities. This framework, termed ConsistencyDet, leverages an innovative denoising concept known as the Consistency Model. The hallmark of this model is its self-consistency feature, which empowers the model to map distorted information from any temporal stage back to its pristine state, thereby realizing a "one-step denoising" mechanism. Such an attribute markedly elevates the operational efficiency of the model, setting it apart from the conventional Diffusion Model. Throughout the training phase, ConsistencyDet initiates the diffusion sequence with noise-infused boxes derived from the ground-truth annotations and conditions the model to perform the denoising task. Subsequently, in the inference stage, the model employs a denoising sampling strategy that commences with bounding boxes randomly sampled from a normal distribution. Through iterative refinement, the model transforms an assortment of arbitrarily generated boxes into definitive detections. Comprehensive evaluations employing standard benchmarks, such as MS-COCO and LVIS, corroborate that ConsistencyDet surpasses other leading-edge detectors in performance metrics. Our code is available at https://***/Tankowa/ConsistencyDet. Copyright © 2024, The Authors. All rights reserved.

关键词： Object detection

data Stream Classification Based on Extreme Learning Machine: A Review

学校读者我要写书评

暂无评论

Big data Research 2022年 30卷

作者： Zheng, Xiulin Li, Peipei Wu, Xindong Key Laboratory of Knowledge Engineering with Big Data (Hefei University of Technology) Ministry of Education China School of Computer Science and Information Engineering Hefei University of Technology Hefei 230601 China Mininglamp Academy of Sciences Mininglamp Technology Beijing 100084 China

Many daily applications are generating massive amount of data in the form of stream at an ever higher speed, such as medical data, clicking stream, internet record and banking transaction, etc. In contrast to the traditional static data, data streams are of some inherent properties, to name a few, infinite length, concept drift, multiple labels and concept evolution. Among all the data mining tasks, classification is one of the basic topics in data stream mining and has gained more and more attentions among different research communities. Extreme Learning Machine (ELM) has drawn much interests in data classification due to its high efficiency, universal approximation capability, generalization ability, and simplicity, which have greatly inspired the development of many ELM-based algorithms and their applications during the past decades. In this paper, we mainly provide a comprehensive review on ELM theoretical research and its variants in data stream classification, and categorize these algorithms from different perspectives. Firstly, we briefly introduce the basic principles of ELM and its characteristics. Secondly, we give an overview of different ELM variants to address the particular issues of data stream classification. Thirdly, we present an overview of different strategies to optimize the ELM, which have further improved the stability, accuracy and generalization ability of ELM, and briefly introduce some practical applications of ELM in data stream classification. Finally, we conduct several groups of experiments to compare the performance of ELM based models addressing the focused issues. Also, the open issues and prospects of ELM models used for stream classification are discussed, which are worthwhile to be further studied in the future. © 2022

关键词： Classification data stream Extreme learning machine

Efficient compressed index for top-k spatial keyword query

学校读者我要写书评

暂无评论

Ruan Jian Xue Bao/Journal of Software 2014年 25卷 157-168页

作者： Zhou, Xin Zhang, Xiao An, Run-Gong Xue, Zhong-Bin Wang, Shan Key Laboratory of Data Engineering and Knowledge Engineering of the Ministry of Education Renmin University of China Beijing100872 China School of Information Renmin University Beijing100872 China

Location-Based services guide a user to find the object which provides services located in a particular position or region (e.g., looking for a coffee shop near a university). Given a query location and multiple keywords, location-based services return the most relevant objects ranked according to location proximity and text relevancy. Various hybrid indexes have been proposed in recent years which combine R-tree and inverted index to improve query efficiency. Unfortunately, the state-of-the-art approaches require more space in order to reduce response time. Cache mechanism is inefficient due to huge storage overhead. In this paper, a novel index based on index compressed technology (CSTI) is proposed, to answer top-k SKQ. CSTI significantly reduces storage overhead (by at least 80%), while maintaining efficient query performance. Extensive experiments based on real dataset and simulated dataset confirm CSTI is effective and efficient. ©2014 ISCAS.

关键词： Location

ViGT: Proposal-free Video Grounding with Learnable Token in Transformer

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Li, Kun Guo, Dan Wang, Meng School of Computer Science and Information Engineering Hefei University of Technology Hefei230601 China Key Laboratory of Knowledge Engineering with Big Data Ministry of Education Hefei230601 China Intelligent Interconnected Systems Laboratory of Anhui Province Hefei230601 China Institute of Artificial Intelligence Hefei Comprehensive National Science Center Hefei230088 China

The video grounding (VG) task aims to locate the queried action or event in an untrimmed video based on rich linguistic descriptions. Existing proposal-free methods are trapped in complex interaction between video and query, overemphasizing cross-modal feature fusion and feature correlation for VG. In this paper, we propose a novel boundary regression paradigm that performs regression token learning in a transformer. Particularly, we present a simple but effective proposal-free framework, namely Video Grounding Transformer (ViGT), which predicts the temporal boundary using a learnable regression token rather than multi-modal or cross-modal features. In ViGT, the benefits of a learnable token are manifested as follows. (1) The token is unrelated to the video or the query and avoids data bias toward the original video and query. (2) The token simultaneously performs global context aggregation from video and query features. First, we employed a sharing feature encoder to project both video and query into a joint feature space before performing cross-modal co-attention (i.e., video-to-query attention and query-to-video attention) to highlight discriminative features in each modality. Furthermore, we concatenated a learnable regression token [REG] with the video and query features as the input of a vision-language transformer. Finally, we utilized the token [REG] to predict the target moment and visual features to constrain the foreground and background probabilities at each timestamp. The proposed ViGT performed well on three public datasets: ANet Captions, TACoS and YouCookII. Extensive ablation studies and qualitative analysis further validated the interpretability of ViGT. Copyright © 2023, The Authors. All rights reserved.

关键词： Regression analysis

Document-level Relation Extraction with Relation Correlations

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Han, Ridong Peng, Tao Wang, Benyou Liu, Lu Wan, Xiang College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education China College of Software Jilin University China Shenzhen Research Institute of Big Data The Chinese University of Hong Kong Shenzhen China School of Data Science The Chinese University of Hong Kong Shenzhen China

Document-level relation extraction faces two overlooked challenges: long-tail problem and multi-label problem. Previous work focuses mainly on obtaining better contextual representations for entity pairs, hardly address the above challenges. In this paper, we analyze the co-occurrence correlation of relations, and introduce it into DocRE task for the first time. We argue that the correlations can not only transfer knowledge between data-rich relations and data-scarce ones to assist in the training of tailed relations, but also reflect semantic distance guiding the classifier to identify semantically close relations for multi-label entity pairs. Specifically, we use relation embedding as a medium, and propose two co-occurrence prediction sub-tasks from both coarse- and fine-grained perspectives to capture relation correlations. Finally, the learned correlation-aware embeddings are used to guide the extraction of relational facts. Substantial experiments on two popular DocRE datasets are conducted, and our method achieves superior results compared to baselines. Insightful analysis also demonstrates the potential of relation correlations to address the above challenges. Copyright © 2022, The Authors. All rights reserved.

关键词： Extraction

Counterfactual Faithful data Generation Based on Disentangled Representation for Compound Fault Diagnosis of Rolling Bearings

学校读者我要写书评

暂无评论

Counterfactual Faithful Data Generation Based on Disentangle...

Sensing, Measurement & data Analytics in the era of Artificial Intelligence (ICSMD), International Conference on

作者： Xu Ding Qiang Zhu Song Wang Yiqi Zhang Hui Wang Juan Xu Institute of Industry and Equipment Technology Hefei University of Technology Hefei China School of Mechanical Engineering Hefei University of Technology Hefei China Bengbu Triumph Engineering Technology Co. Ltd Bengbu China Key Laboratory of Knowledge Engineering with Big Data School of Computer and Information Hefei University of Technology Hefei China

Recently, deep learning and human-out-of-the-loop methods enjoy their prosperous applications in mechanical fault diagnosis. Nonetheless, the None-IID(independent and identically distributed) issue radicated in acquired data severely limits the stability and accuracy of compound fault diagnosis of rolling bearings. This paper proposes a sample augmentation method for generating simulated signals based on the concept of counterfactuals. Firstly, disentangled representations and counterfactual faithful theory are applied to classify the original signal into two categories of properties. One is the fault semantics encoded from the original vibration signal. And the other is the sample attribute encoded by the encoder of Variational Autoencoders (VAEs). Secondly, the counterfactual faithful pseudo-samples are conjured through the Generative Adversarial Network(GAN) using the seeds of the “factual” sample attributes and “counterfactual” fault semantics to compensate for the drawback of distribution shift. Finally, the original samples and pseudo-samples are used as the CNN classifier dataset to realize bearing fault diagnosis. Experiments show that this method can generate counterfactual signals that are highly consistent with the original data distribution and can achieve better classification accuracy after balancing the dataset.

关键词： Fault diagnosis Wavelet transforms Vibrations Semantics Rolling bearings Feature extraction Stability analysis

Computationally Efficient Approximations for Matrix-based Rényi's Entropy

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Gong, Tieliang Dong, Yuxin Yu, Shujian Dong, Bo The School of Computer Science and Technology Xi'an Jiaotong University Shaanxi Provincial Key Laboratory of Big Data Knowledge Engineering Xi’an710049 China The Machine Learning Group UiT - The Arctic University of Norway Department of Computer Science Vrije University Amsterdam Amsterdam Netherlands

The recently developed matrix-based Rényi's αorder entropy enables measurement of information in data simply using the eigenspectrum of symmetric positive semi-definite (PSD) matrices in reproducing kernel Hilbert space, without estimation of the underlying data distribution. This intriguing property makes this new information measurement widely adopted in multiple statistical inference and learning tasks. However, the computation of such quantity involves the trace operator on a PSD matrix G to power α (i.e., tr(Gα)), with a normal complexity of nearly O(n3), which severely hampers its practical usage when the number of samples (i.e., n) is large. In this work, we present computationally efficient approximations to this new entropy functional that can reduce its complexity to even significantly less than O(n2). To this end, we leverage the recent progress on Randomized Numerical Linear Algebra, developing Taylor, Chebyshev and Lanczos approximations to tr(Gα) for arbitrary values of α by converting it into a matrix-vector multiplication problem. We also establish the connection between the matrix-based Rényi's entropy and PSD matrix approximation, which enables exploiting both clustering and block low-rank structure of G to further reduce the computational cost. We theoretically provide approximation accuracy guarantees and illustrate the properties for different approximations. Large-scale experimental evaluations on both synthetic and real-world data corroborate our theoretical findings, showing promising speedup with negligible loss in accuracy. Copyright © 2021, The Authors. All rights reserved.

关键词： Matrix algebra