Deep Generative Model for Multi-Class Imbalanced Learning

Zhang, Yazhou

doi:10.23860/thesis-zhang-yazhou-2018

Cited by 5 publications

(10 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By leveraging the synthetic sample generation capabilities of GANs. The theoretical foundation and implementation strategies of this GAN-based balancing approach, highlighting its potential to improve classification performance on all classes [1,30].…”

Section: Gans Based Balancingmentioning

confidence: 99%

“…Triple-GANs (TGAN) is a three character architecture implemented to solve the problem of imbalanced dataset using oversampling technique. It uses an adversarial relationship between classifier (a), generator (g) and discriminator (d) [30]. Though the generation process is benefited but the classification is unable to learn the things generated due to adversarial relationship.…”

Section: Gans Based Balancingmentioning

confidence: 99%

See 1 more Smart Citation

Enhancing Classification of Imbalanced Data using Diffusion Process

Velankar,

Patil

2024

Preprint

View full text Add to dashboard Cite

Conventional deep learning algorithms have difficulties when imbalanced data is encountered where one class disproportionately outnumbers another. This work investigates how a novel deep learning approach called diffusion processes could enhance classification performance on unbalanced datasets. We study how well diffusion models can produce artificial information for underrepresented group, equalizing the spread of categories and reducing innate biases in favour of the majority class. The study also looks at how various diffusion model topologies and training approaches affect classification accuracy, with a focus on minority classes. Furthermore, the study intends to evaluate diffusion-based approaches against other cutting-edge strategies for managing unbalanced data. The results of this study should help build more reliable and accurate Deep Learning models for practical uses, as well as offer insightful information on how well diffusion processes work for the generation of unbalanced data.

show abstract

Section: Gans Based Balancingmentioning

confidence: 99%

Section: Gans Based Balancingmentioning

confidence: 99%

Enhancing Classification of Imbalanced Data using Diffusion Process

Velankar,

Patil

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Other methods. Other paradigms include task-specific architecture design (Wang et al, 2021c;Zhou et al, 2020;Wang et al, 2021a), transfer learning (Liu et al, 2019;Yin et al, 2019), domain adaptation (Jamal et al, 2020), semisupervised learning, and self-supervised learning (Yang and Xu, 2020) which demand nontrivial architecture design or external data.…”

Section: Imbalanced Learningmentioning

confidence: 99%

“…The zero-shot classification performance of VLMs on imbalanced datasets is limited due to several factors, including the inherent bias in the pre-training data, the lack of exposure to the tail classes during pre-training, and the lack of techniques to mitigate the effects of class imbalance (Schuhmann et al, 2022). As a result, VLMs often perform poorly on tail classes, which can be critical for many applications related to safety or health, such as autonomous driving and medical diagnosis (Yang and Xu, 2020). Hence, it is intuitive to ask: is this long-tailed pre-training style actually transfer or influence downstream long-tailed classes?…”

Section: Introductionmentioning

confidence: 99%

Linguistic Semiotics

Wang¹

2020

Peking University Linguistics Research

View full text Add to dashboard Cite

Vision-Language models (VLMs) that use contrastive language-image pre-training have shown promising zero-shot classification performance. However, their performance on imbalanced dataset is relatively poor, where the distribution of classes in the training dataset is skewed, leading to poor performance in predicting minority classes. For instance, CLIP achieved only 5% accuracy on the iNaturalist18 dataset. We propose to add a lightweight decoder to VLMs to avoid OOM (out of memory) problem caused by large number of classes and capture nuanced features for tail classes. Then, we explore improvements of VLMs using prompt tuning, fine-tuning, and incorporating imbalanced algorithms such as Focal Loss, Balanced SoftMax and Distribution Alignment. Experiments demonstrate that the performance of VLMs can be further boosted when used with decoder and imbalanced methods. Specifically, our improved VLMs significantly outperforms zero-shot classification by an average accuracy of 6.58%, 69.82%, and 6.17%, on ImageNet-LT, iNaturalist18, and Places-LT, respectively. We further analyze the influence of pre-training data size, backbones, and training cost. Our study highlights the significance of imbalanced learning algorithms in face of VLMs pre-trained by huge data. We release our code at https://github.com/Imbalance-VLM/Imbalance-VLM.

show abstract

“…However, standard datasets available for such tasks are typically of high dimensionality and unbalanced, where each class label is not equally represented. This imbalance can potentially impact the performance of the classifier [8,9].…”

Section: Introductionmentioning

confidence: 99%

Evaluating the Efficacy of Resampling Techniques in Addressing Class Imbalance for Network Intrusion Detection Systems Using Support Vector Machines

Kudithipudi,

Narisetty,

Kancherla

et al. 2023

ISI

View full text Add to dashboard Cite

The objective of this study was to assess the performance of various resampling strategies aimed at mitigating the class imbalance problem in Network Intrusion Detection Systems (NIDS) using machine learning models and imbalanced benchmark datasets. Due to this class imbalance problem, detection of known or unknown attacks in NIDS often results in suboptimal performance. Resampling methods, statistically designed to generate synthetic samples from existing datasets, were employed to rebalance class labels and train the machine learning models. The Support Vector Machine (SVM), a robust supervised classifier, was utilized to classify data by identifying the optimal decision boundary that maximally separates different classes. In this context, efforts were made to enhance the effectiveness of these resampling techniques and consider the potential benefits of hybrid models. No resampling (NR), Synthetic Minority Over-sampling Technique (SMOTE), Random Under Sampling (RUS), Random Under Sampling and Random Over Sampling (RUS+ROS), and Random Under Sampling and SMOTE (RUS+SMOTE) were evaluated. The SVM classifier with Radial Basis Function (RBF) was employed, validated against the imbalanced benchmark dataset CICIDS-2017 (Canadian Institute for Cyber Security Intrusion Detection dataset-2017), to assess the effectiveness of these methods using performance metrics such as Accuracy, Precision, Recall, F1 score, and wall time. The proposed method achieved a remarkable accuracy of 99.63% in intrusion detection, demonstrating impressive results when compared to state-of-the-art methods for detecting network attacks on imbalanced datasets. The findings from this research provide valuable insights into the potential of various resampling methods in tackling class imbalance problems in NIDS.

show abstract

Deep Generative Model for Multi-Class Imbalanced Learning

Cited by 5 publications

References 41 publications

Enhancing Classification of Imbalanced Data using Diffusion Process

Enhancing Classification of Imbalanced Data using Diffusion Process

Linguistic Semiotics

Evaluating the Efficacy of Resampling Techniques in Addressing Class Imbalance for Network Intrusion Detection Systems Using Support Vector Machines

Contact Info

Product

Resources

About