检索结果-内蒙古大学图书馆

Targeting Negative Flips in Active Learning using Validation Sets

学校读者我要写书评

暂无评论

Targeting Negative Flips in Active Learning using Validation...

IEEE International Conference on Big Data

作者： Ryan Benkert Mohit Prabhushankar Ghassan AlRegib OLIVES at the Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology Atlanta GA USA

ISBN: (数字)9798350362480

ISBN: (纸本)9798350362497

The performance of active learning algorithms can be improved in two ways. The often used and intuitive way is by reducing the overall error rate within the test set. The second way is to ensure that correct predictions are not forgotten when the training set is increased in between rounds. The former is measured by the accuracy of the model and the latter is captured in negative flips between rounds. Negative flips are samples that are correctly predicted when trained with the previous/smaller dataset and incorrectly predicted after additional samples are labeled. In this paper, we discuss improving the performance of active learning algorithms both in terms of prediction accuracy and negative flips. The first observation we make in this paper is that negative flips and overall error rates are decoupled and reducing one does not necessarily imply that the other is reduced. Our observation is important as current active learning algorithms do not consider negative flips directly and implicitly assume the opposite. The second observation is that performing targeted active learning on subsets of the unlabeled pool has a significant impact on the behavior of the active learning algorithm and influences both negative flips and prediction accuracy. We then develop ROSE - a plug-in algorithm that utilizes a small labeled validation set to restrict arbitrary active learning acquisition functions to negative flips within the unlabeled pool. We show that integrating a validation set results in a significant performance boost in terms of accuracy, negative flip rate reduction, or both.

关键词： Training Accuracy Spirals Error analysis Active learning Big Data Prediction algorithms Software

Targeting Negative Flips in Active Learning using Validation Sets

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Benkert, Ryan Prabhushankar, Mohit AlRegib, Ghassan OLIVES The Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA United States

关键词： Active learning

Taxes Are All You Need: Integration of Taxonomical Hierarchy Relationships into the Contrastive Loss

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Kokilepersaud, Kiran Yarici, Yavuz Prabhushankar, Mohit AlRegib, Ghassan OLIVES The Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA United States

In this work, we propose a novel supervised contrastive loss that enables the integration of taxonomic hierarchy information during the representation learning process. A supervised contrastive loss operates by enforcing that images with the same class label (positive samples) project closer to each other than images with differing class labels (negative samples). The advantage of this approach is that it directly penalizes the structure of the representation space itself. This enables greater flexibility with respect to encoding semantic concepts. However, the standard supervised contrastive loss only enforces semantic structure based on the downstream task (i.e. the class label). In reality, the class label is only one level of a hierarchy of different semantic relationships known as a taxonomy. For example, the class label is oftentimes the species of an animal, but between different classes there are higher order relationships such as all animals with wings being "birds". We show that by explicitly accounting for these relationships with a weighting penalty in the contrastive loss we can out-perform the supervised contrastive loss. Additionally, we demonstrate the adaptability of the notion of a taxonomy by integrating our loss into medical and noise-based settings that show performance improvements by as much as 7%. © 2024, CC BY.

关键词： Taxonomies

Explaining Representation Learning with Perceptual Components

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Yarici, Yavuz Kokilepersaud, Kiran Prabhushankar, Mohit AlRegib, Ghassan OLIVES The Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA United States

Self-supervised models create representation spaces that lack clear semantic meaning. This interpretability problem of representations makes traditional explainability methods ineffective in this context. In this paper, we introduce a novel method to analyze representation spaces using three key perceptual components: color, shape, and texture. We employ selective masking of these components to observe changes in representations, resulting in distinct importance maps for each. In scenarios, where labels are absent, these importance maps provide more intuitive explanations as they are integral to the human visual system. Our approach enhances the interpretability of the representation space, offering explanations that resonate with human visual perception. We analyze how different training objectives create distinct representation spaces using perceptual components. Additionally, we examine the representation of images across diverse image domains, providing insights into the role of these components in different contexts. © 2024, CC BY.

关键词： Textures

HEX: Hierarchical Emergence Exploitation in Self-Supervised Algorithms

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Kokilepersaud, Kiran Kim, Seulgi Prabhushankar, Mohit AlRegib, Ghassan OLIVES The Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA United States

In this paper, we propose an algorithm that can be used on top of a wide variety of self-supervised (SSL) approaches to take advantage of hierarchical structures that emerge during training. SSL approaches typically work through some invariance term to ensure consistency between similar samples and a regularization term to prevent global dimensional collapse. Dimensional collapse refers to data representations spanning a lower-dimensional subspace. Recent work has demonstrated that the representation space of these algorithms gradually reflects a semantic hierarchical structure as training progresses. Data samples of the same hierarchical grouping tend to exhibit greater dimensional collapse locally compared to the dataset as a whole due to sharing features in common with each other. Ideally, SSL algorithms would take advantage of this hierarchical emergence to have an additional regularization term to account for this local dimensional collapse effect. However, the construction of existing SSL algorithms does not account for this property. To address this, we propose an adaptive algorithm that performs a weighted decomposition of the denominator of the InfoNCE loss into two terms: local hierarchical and global collapse regularization respectively. This decomposition is based on an adaptive threshold that gradually lowers to reflect the emerging hierarchical structure of the representation space throughout training. It is based on an analysis of the cosine similarity distribution of samples in a batch. We demonstrate that this hierarchical emergence exploitation (HEX) approach can be integrated across a wide variety of SSL algorithms. Empirically, we show performance improvements of up to 5.6% relative improvement over baseline SSL approaches on classification accuracy on imagenet with 100 epochs of training. © 2024, CC BY.

关键词： Consensus algorithm

INTELLIGENT MULTI-VIEW TEST TIME AUGMENTATION

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Ozturk, Efe Prabhushankar, Mohit AlRegib, Ghassan OLIVES The Center for Signal and Information Processing CSIP School of Electrical and Computer Engineering Georgia Institute of Technology AtlantaGA United States

In this study, we introduce an intelligent Test Time Augmentation (TTA) algorithm designed to enhance the robustness and accuracy of image classification models against viewpoint variations. Unlike traditional TTA methods that indiscriminately apply augmentations, our approach intelligently selects optimal augmentations based on predictive uncertainty metrics. This selection is achieved via a two-stage process: the first stage identifies the optimal augmentation for each class by evaluating uncertainty levels, while the second stage implements an uncertainty threshold to determine when applying TTA would be advantageous. This methodological advancement ensures that augmentations contribute to classification more effectively than a uniform application across the dataset. Experimental validation across several datasets and neural network architectures validates our approach, yielding an average accuracy improvement of 1.73% over methods that use single-view images. This research underscores the potential of adaptive, uncertainty-aware TTA in improving the robustness of image classification in the presence of viewpoint variations, paving the way for further exploration into intelligent augmentation strategies. The code is available at: https://***/olivesgatech/Intelligent-Multi-View-TTA © 2024, CC BY.

关键词： Deep neural networks

Multimedia Classification via Tensor Linear Discriminant Analysis

学校读者我要写书评

暂无评论

IEEE Transactions on Broadcasting 2024年第4期70卷 1139-1152页

作者： Chang, Shih-Yu Wu, Hsiao-Chun Yan, Kun Chih-Hao Huang, Scott Wu, Yiyan San Jose State University Department Of Applied Data Science San JoseCA95192 United States Louisiana State University School Of Electrical Engineering And Computer Science Baton RougeLA70803 United States Yuan Ze University Innovation Center For Ai Applications Taoyuan32003 Taiwan Guilin University Of Electronic Technology Guangxi Key Laboratory Of Wireless Wideband Communication And Signal Processing Department Of Information And Telecommunication Guilin541004 China National Tsing Hua University Department Of Electrical Engineering Hsinchu300 Taiwan Western University Department Of Electrical And Computer Engineering LondonONN6A 3K7 Canada

Linear discriminant analysis (LDA) is a well-known feature-extraction technique for data analytic and pattern classification. As the dimensionality of multimedia data has increased in this big era, it is often to characterize data by tensors. Over the past two decades, researchers have thus explored to extend LDA to the general tensor space, especially in two common ways: LDA of tensors using tensor decomposition methods (by conversion of tensors to matrices) and LDA of tensors built upon the T-product. However, both of the aforementioned approaches have restrictions thereby. A critical problem about how to carry out LDA of arbitrary scatter tensors based on the Einstein product still remains unsolved by the existing methods. Therefore, we propose a novel tensor LDA (a.k.a. TLDA) approach, which can carry out the LDA of arbitrary-dimensional scatter-tensors without any need of tensor decomposition. Besides, for reducing the computation time, we also design a parallel paradigm to execute our proposed TLDA in this work. Numerical experiments conducted over real multimedia data demonstrate the efficacy of our proposed new TLDA in terms of classification accuracy. Moreover, the comparison of the classification accuracies, computational-complexities, and memory-complexities of our proposed novel TLDA scheme and other existing tensor-based LDA methods is made. By leveraging TLDA for high-dimensional feature extraction, segmentation, and user-item interaction data processing, future multimedia recommendation systems can facilitate more accurate, engaging, and satisfactory user experience over the Internet. © 1963-12012 IEEE.

关键词： Tensors

Complementary Phase Encoding for Pair-Wise Neural Deblurring of Accelerated Brain MRI 17th

学校读者我要写书评

暂无评论

Complementary Phase Encoding for Pair-Wise Neural Deblurrin...

17th European Conference on computer Vision, ECCV 2022

作者： Hod, Gali Green, Michael Waserman, Mark Konen, Eli Shrot, Shai Nelkenbaum, Ilya Kiryati, Nahum Mayer, Arnaldo School of Electrical Engineering Tel Aviv University Tel Aviv Israel School of Computer Sciences Ben Gurion University Beersheba Israel Diagnostic Imaging Department at Sheba Medical Center Affiliated with the Beverly Sackler School of Medicine Tel Aviv University Tel Aviv Israel Klachky Chair of Image Processing School of Electrical Engineering Tel Aviv University Tel Aviv Israel

ISBN: (纸本)9783031250651

MRI has become an invaluable tool for diagnostic brain imaging, providing unrivalled qualitative and quantitative information to the radiologist. However, due to long scanning times and capital costs, access to MRI lags behind CT. Typical brain protocols lasting over 30 min set a clear limitation to patient experience, scanner throughput, operation profitability, and lead to long waiting times for an appointment. As image quality, in terms of spatial resolution and noise, is strongly dependent on acquisition duration, significant scanning acceleration must successfully address challenging image degradation. In this work, we consider the scan acceleration scenario of a strongly anisotropic acquisition matrix. We propose a neural approach that jointly deblurs scan pairs acquired with mutually orthogonal phase encoding directions. This leverages the complementarity of the respective phase encoded information as blur directions are also mutually orthogonal between the scans in the pair. The proposed architecture, trained end-to-end, is applied to T1w scan pairs consisting of one scan with contrast media injection (CMI), and one without. Qualitative and quantitative validation is provided against state-of-the-art deblurring methods, for an acceleration factor of 4 beyond compressed sensing acceleration. The proposed method outperforms the compared methods, suggesting its possible clinical applicability for this challenging task. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Magnetic resonance imaging

Imperceptible and Sparse Adversarial Attacks via a Dual-Population-Based Constrained Evolutionary Algorithm

学校读者我要写书评

暂无评论

IEEE Transactions on Artificial Intelligence

IEEE Transactions on Artificial Intelligence 2023年第2期4卷 268-281页

作者： Tian, Ye Pan, Jingwen Yang, Shangshang Zhang, Xingyi He, Shuping Jin, Yaochu Anhui University Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education Institutes of Physical Science and Information Technology Hefei230601 China Hefei Comprehensive National Science Center Institute of Artificial Intelligence Hefei230088 China Anhui University School of Computer Science and Technology Hefei230601 China Anhui University Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education School of Artificial Intelligence Hefei230601 China Anhui University Anhui Engineering Laboratory of Human-Robot Integration System and Intelligent Equipment School of Electrical Engineering and Automation Hefei230601 China Bielefeld University Faculty of Technology Bielefeld33619 Germany

The sparse adversarial attack has attracted increasing attention due to the merit of a low attack cost via changing a small number of pixels. However, the generated adversarial examples are easily detected in vision since the perturbation to each pixel is relatively large. To achieve imperceptible and sparse adversarial attacks, this article formulates a bi-objective constrained optimization problem, simultaneously minimizing the 0 and 2 distances to the original image, and proposes a dual-population-based constrained evolutionary algorithm to solve it. The proposed method solves the optimization problem by evolving two populations, where one population is responsible for finding feasible solutions (i.e., successful attacks) and the other one is to minimize both the 0 and 2 distances. Moreover, a population initialization strategy and two genetic operators are customized to accelerate the convergence speed. Experimental results indicate that the proposed method can achieve high success rates with low attack costs, and strikes a better balance between the 0 and 2 distances than state-of-the-art methods. © 2020 IEEE.

关键词： Evolutionary algorithms