Deep Self-Learning From Noisy Labels

Han, Jie Cai; Luo, Ping; Wang, Xiaogang

doi:10.1109/iccv.2019.00524

Cited by 292 publications

(212 citation statements)

References 31 publications

Supporting

Mentioning

196

Contrasting

Order By: Relevance

“…Some structures, such as the infundibula, might otherwise be misdiagnosed as aneurysms. Unambiguous interpretations, which are called noisy labels can lead to an obviously biased performance of the model 39,40 . Therefore, it is necessary and urgent to use DSA, the gold standard for IAs detection in machine learning based CAD studies.…”

Section: Discussionmentioning

confidence: 99%

A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images

et al. 2020

View full text Add to dashboard Cite

Intracranial aneurysm is a common life-threatening disease. Computed tomography angiography is recommended as the standard diagnosis tool; yet, interpretation can be time-consuming and challenging. We present a specific deep-learning-based model trained on 1,177 digital subtraction angiography verified bone-removal computed tomography angiography cases. The model has good tolerance to image quality and is tested with different manufacturers. Simulated real-world studies are conducted in consecutive internal and external cohorts, in which it achieves an improved patient-level sensitivity and lesion-level sensitivity compared to that of radiologists and expert neurosurgeons. A specific cohort of suspected acute ischemic stroke is employed and it is found that 99.0% predicted-negative cases can be trusted with high confidence, leading to a potential reduction in human workload. A prospective study is warranted to determine whether the algorithm could improve patients’ care in comparison to clinicians’ assessment.

show abstract

Section: Discussionmentioning

confidence: 99%

A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images

et al. 2020

View full text Add to dashboard Cite

show abstract

“…The model learns the features of each category in a sample set well when the number of samples is small. Then the first batch of 5000 samples are labeled according to the initial model, the appropriate photos whose recognition accuracy is higher than the accuracy threshold [32,33](70%) and artificially selected are put into the candidate photos which are added to the next batch of training samples after preprocessing. If the accuracy threshold is too low, the noise of features of non-class samples will confuse the category of samples and increase the workload of artificial verification.…”

Section: ) Methods To Construct the Datasetmentioning

confidence: 99%

A Framework for Land Use Scenes Classification Based on Landscape Photos

Zhang

Zeng

et al. 2020

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Space-Earth Integrated Stereoscopic Mapping promotes the progress of earth observation technologies. The method which combined remote sensing images with zenith perspectives and ground-level landscape photos with slanted viewing angles improves the efficiency and accuracy of land surveys. Recently, numerous efforts have been devoted to combining deep learning and remote sensing images for the classification of land use scenes. However, improvement of classification accuracy has been limited because of the lack of sectional representation. Landscape photos can describe the crosssections in detail. For this reason, this study constructed a Landuse Semantic Photo Dataset (LSPD) and proposed a Land-use Classification Framework for Photos (LUCFP) based on Inception-v4. LSPD was constructed through semantic planning, scene segmentation, supervised iteration transfer learning and augmentation of photos. LSPD has 1.4 million photos collected from seven geographic regions of China, and covers 13 land-use categories and 44 semantic categories. LUCFP adapts scene segmentation based on depth of field, multi-semantic block labeling, and weighting of semantic joint spatial ranges to determine the land use category. To validate LUCFP, nine semantic samples (9×3×2000 photos) were chosen from LSPD, obtaining an overall accuracy of 97.64%. The best photo cropping method was masking, which crops the boundary of the scene labeled by the photo, leading to an accuracy of 90.32%. The optimal pixel size that balances speed and accuracy is 675×675, with speed reaching 30 photos per second with an average accuracy of 93.80%. LUCFP has been successfully applied to the automatic verification of land surveys in China. Index Terms-deep convolutional neural networks (DCNN), landscape photos, land survey, land use scene classification I. INTRODUCTION and use classification is essential for applications such as land resource management, urban planning, precision agriculture and environmental protection [1, 2], whose essence is to classify the images (remote sensing images or ground-level landscape photos) that reflect the present situations of land use.

show abstract

“…The proposition of this paper is to utilize publicly available resources and employ Machine and Deep Learning techniques to create a dataset that can be made available for pharmacovigilance research. We believe that we can't keep training models with very limited amount of manually annotated tweets, but we can use the theory of noisy labeling to create more robust models with silver standards [31][32][33] . However, there are a few limitations which we would like to address in our future work.…”

Section: Future Workmentioning

confidence: 99%

Mining Archive.org’s Twitter Stream Grab for Pharmacovigilance Research Gold

Tekumalla

Asl

Banda

2019

Preprint

View full text Add to dashboard Cite

In the last few years Twitter has become an important resource for the identification of Adverse Drug Reactions (ADRs), monitoring flu trends, and other pharmacovigilance and general research applications. Most researchers spend their time crawling Twitter, buying expensive pre-mined datasets, or tediously and slowly building datasets using the limited Twitter API. However, there are a large number of datasets that are publicly available to researchers which are underutilized or unused. In this work, we demonstrate how we mined over 9.4 billion Tweets from archive.org's Twitter stream grab using a drug-term dictionary and plenty of computing power. Knowing that not everything that shines is gold, we used pre-existing drug-related datasets to build machine learning models to filter our findings for relevance. In this work we present our methodology and the 3,346,758 identified tweets for public use in future research.

show abstract

Deep Self-Learning From Noisy Labels

Cited by 292 publications

References 31 publications

A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images

A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images

A Framework for Land Use Scenes Classification Based on Landscape Photos

Mining Archive.org’s Twitter Stream Grab for Pharmacovigilance Research Gold

Contact Info

Product

Resources

About