Home › Categories › Business & Finance› Compounded Corruption: A Fast And Effective Data Augmentation Technique For Out-of-Distribution Sample Detection

Compounded Corruption: A Fast And Effective Data Augmentation Technique For Out-of-Distribution Sample Detection.pdf

2207.13916.pdf
Preview of Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection
🔗 Source: arxiv.org
📊 Size: 4.72 MB
📄 Pages: 43 pages
⬇️ Downloads: 883

Summary

The primary goal is to improve the accuracy and efficiency of OOD detectors, which misclassify OOD samples as in-distribution (ID) classes with high confidence.

Key Advantages of CnC:

1. No Hold-out Data: Unlike most techniques, CnC doesn't require a separate OOD dataset for training.
2. Faster Inference: It avoids backpropagation and ensembling at test time, making it significantly faster during inference.
3. Improved Detection Accuracy: Extensive comparisons show that models trained with CnC-augmented data outperform state-of-the-art (SOTA) methods in terms of OOD detection accuracy and inference speed.
4. Theoretical Insights: Analysis reveals that CnC leads to a tighter boundary around ID classes, facilitating better OOD sample detection.

Comparison with Existing Techniques:

- Different Approach: Unlike techniques using natural corruptions or adversarial perturbations, CnC focuses on generating OOD samples in two key regions: near decision boundaries and uniformly distributed in inter-ID sample space.
- Performance: CnC achieves state-of-the-art results on various datasets, outperforming methods like GODIN by 18.4% in TNR@TPR95.

Key Contributions:

1. Novel data augmentation strategy (CnC).
2. Classification of OOD samples into a separate "reject" class.
3. Faster inference without input pre-processing or additional forward passes.
4. Empirical and theoretical analysis highlighting the importance of polyhedral complex granularity, entropy, and diversity in OOD detection performance.

Download File

Description

A novel data augmentation technique, Compounded Corruptions (CnC), creates synthetic out-of-distribution samples by corrupting in-distribution data, enabling the training of effective OOD detection classifiers without requiring labeled OOD examples.

Technical Information

  • File Format: PDF
  • File Size: 4.72 MB
  • Pages: 43
  • Language: EN
  • Total Downloads: 883
  • Last Updated: 3 weeks ago

Document Overview

This PDF document about Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection provides comprehensive information and guidance. Whether you're a beginner or advanced user, this resource offers valuable insights into Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection.

Related Topics

If you're interested in Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection, you might also want to explore:

Compounded guideCompounded tutorialCorruption: guideCorruption: tutorialFast guideFast tutorialEffective guideEffective tutorialData guideData tutorialAugmentation guideAugmentation tutorialTechnique guideTechnique tutorialOut-of-Distribution guideOut-of-Distribution tutorialSample guideSample tutorialDetection guideDetection tutorial

Related Categories

Education & Learning
22711 documents
Health & Wellness
8956 documents
Technology
4783 documents

Related Documents

Preview of Data Corruption
PDF file

Data Corruption

2.88 MBpascoelPT7 pages
Não foi possível resumir o texto devido ao seu conteúdo aparentemente aleatório e sem significado.
Download
Preview of Data Corruption
PDF file

Data Corruption

294 KBKnobelPT7 pages
O texto fornecido parece ser um conjunto de caracteres aleatórios, símbolos e palavras desconexas, sem um padrão ou significado claro.
Download
Preview of Deception Detection
PDF file

Deception Detection

226 KBEN4 pages
Researchers proposed a negotiation dialog management system that detects user lies and reacts accordingly, using a doctor-patient conversation model...
Download
Preview of Out-of-Distribution Generalization
PDF file

Out-of-Distribution Generalization

2.64 MBEN30 pages
Dinghuai Zhang discusses "spurious" correlation, "true label" and "spurious label", and references GroupDRO and the paper "DISTRIBUTIONALLY ROBUST...
Download
Preview of PDF
PDF file

PDF

184 KBomicsEN2 pages
Viral infections are a major cause of morbidity and mortality worldwide, and rapid and accurate diagnosis is essential for timely treatment and...
Download

Download Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection eBooks for free and learn more about Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection. These books contain exercises and tutorials to improve your practical skills, at all levels!

Not satisfied with this document? We have related documents to Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection, try searching with similar keywords: Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection, Självkänsla och KBT ‐ Påverkas självkänslan vid KBT för depression och ångesttillstånd?Se lf‐esteem and CBT ‐ How does CBT for de, "Hinder och möjliggörare för 1.5°-livsstilar: Ytliga och djupgående strukturella faktorer som påverkar potentialen för hållbar k, Ändring av genomföranderam för en europeisk plattform för utbyte av balansenergi från frekvensåterställn ingsreserver med manuell, Rekommendationer för vaccination mot covid-19 för särskilda grupper av barn -, förstudie för att utvärdera förutsättningarna att genom en innovationsupphandli ng utveckla en drifttjänst för geoenergilager, Matglädje för alla: en guide till rätt konsistens för olika behov, AUGMENTATION Augmentation Core Rulebook MSRP Pages

You can download PDF versions of the user's guide, manuals and ebooks about Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection, you can also find and download for free A free online manual (notices) with beginner and intermediate, Downloads Documentation, You can download PDF files (or DOC and PPT) about Compounded Corruption: A Fast and Effective Data Augmentation Technique for Out-of-Distribution Sample Detection for free, but please respect copyrighted ebooks.