检索结果-内蒙古大学图书馆

Analyzing the Structure of Attention in a Transformer Language Model

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Vig, Jesse Belinkov, Yonatan Palo Alto Research Center Machine Learning and Data Science Group Interaction and Analytics Lab Palo AltoCA United States Harvard John A. Paulson School of Engineering and Applied Sciences MIT Computer Science and Artificial Intelligence Laboratory CambridgeMA United States

The Transformer is a fully attention-based alternative to recurrent networks that has achieved state-of-the-art results across a range of NLP tasks. In this paper, we analyze the structure of attention in a Transformer language model, the GPT-2 small pretrained model. We visualize attention for individual instances and analyze the interaction between attention and syntax over a large corpus. We find that attention targets different parts of speech at different layer depths within the model, and that attention aligns with dependency relations most strongly in the middle layers. We also find that the deepest layers of the model capture the most distant relationships. Finally, we extract exemplar sentences that reveal highly specific patterns targeted by particular attention heads. Copyright © 2019, The Authors. All rights reserved.

关键词： Recurrent neural networks

Why is the winner the best?

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Eisenmann, M. Reinke, A. Weru, V. Tizabi, M.D. Isensee, F. Adler, T.J. Ali, S. Andrearczyk, V. Aubreville, M. Baid, U. Bakas, S. Balu, N. Bano, S. Bernal, J. Bodenstedt, S. Casella, A. Cheplygina, V. Daum, M. de Bruijne, M. Depeursinge, A. Dorent, R. Egger, J. Ellis, D.G. Engelhardt, S. Ganz, M. Ghatwary, N. Girard, G. Godau, P. Gupta, A. Hansen, L. Harada, K. Heinrich, M. Heller, N. Hering, A. Huaulmé, A. Jannin, P. Kavur, A.E. Kodym, O. Kozubek, M. Li, J. Li, H. Ma, J. Martín-Isla, C. Menze, B. Noble, A. Oreiller, V. Padoy, N. Pati, S. Payette, K. Rädsch, T. Rafael-Patiño, J. Bawa, V. Singh Speidel, S. Sudre, C.H. van Wijnen, K. Wagner, M. Wei, D. Yamlahi, A. Yap, M.H. Yuan, C. Zenk, M. Zia, A. Zimmerer, D. Aydogan, D. Bhattarai, B. Bloch, L. Brüngel, R. Cho, J. Choi, C. Dou, Q. Ezhov, I. Friedrich, C.M. Fuller, C. Gaire, R.R. Galdran, A. Faura, Á. García Grammatikopoulou, M. Hong, S. Jahanifar, M. Jang, I. Kadkhodamohammadi, A. Kang, I. Kofler, F. Kondo, S. Kuijf, H. Li, M. Luu, M. Martinčič, T. Morais, P. Naser, M.A. Oliveira, B. Owen, D. Pang, S. Park, J. Park, S. Plotka, S. Puybareau, E. Rajpoot, N. Ryu, K. Saeed, N. Shephard, A. Shi, P. Štepec, D. Subedi, R. Tochon, G. Torres, H.R. Urien, H. Vilaça, J.L. Wahid, K.A. Wang, H. Wang, J. Wang, L. Wang, X. Wiestler, B. Wodzinski, M. Xia, F. Xie, J. Xiong, Z. Yang, S. Yang, Y. Zhao, Z. Maier-Hein, K. Jäger, P.F. Kopp-Schneider, A. Maier-Hein, L. Heidelberg Germany Heidelberg Germany Faculty of Mathematics and Computer Science Heidelberg University Heidelberg Germany Heidelberg Germany Heidelberg Germany School of Computing Faculty of Engineering and Physical Sciences University of Leeds Leeds United Kingdom Institute of Informatics School of Management HES-SO Valais-Wallis University of Applied Sciences and Arts Western Switzerland Sierre Switzerland Department of Nuclear Medicine and Molecular Imaging Lausanne University Hospital Lausanne Switzerland Technische Hochschule Ingolstadt Ingolstadt Germany University of Pennsylvania PhiladelphiaPA United States Department of Pathology and Laboratory Medicine Perelman School of Medicine University of Pennsylvania PhiladelphiaPA United States Department of Radiology Perelman School of Medicine University of Pennsylvania PhiladelphiaPA United States Department of Radiology University of Washington SeattleWA United States Department of Computer Science University College London London United Kingdom Universitat Autònoma de Barcelona & Computer Vision Center Barcelona Spain Dresden Dresden Germany Department of Advanced Robotics Istituto Italiano di Tecnologia Italy IT University of Copenhagen Copenhagen Denmark Department of General Visceral and Transplantation Surgery Heidelberg University Hospital Heidelberg Germany Biomedical Imaging Group Rotterdam Department of Radiology and Nuclear Medicine Erasmus MC Rotterdam Netherlands Department of Computer Science University of Copenhagen Copenhagen Denmark Sierre Switzerland Harvard Medical School Brigham and Women’s Hospital BostonMA United States School of Biomedical Engineering and Imaging Sciences King’s College London London United Kingdom Essen Germany University of Nebraska Medical Center OmahaNE United States Department of Internal Medicine III Heidelberg University Hospital Heidelberg Germany Neurobiology Research Unit Copenhagen University Hospital Rigshospitalet C

International benchmarking competitions have become fundamental for the comparative performance assessment of image analysis methods. However, little attention has been given to investigating what can be learnt from these competitions. Do they really generate scientific progress? What are common and successful participation strategies? What makes a solution superior to a competing method? To address this gap in the literature, we performed a multicenter study with all 80 competitions that were conducted in the scope of IEEE ISBI 2021 and MICCAI 2021. Statistical analyses performed based on comprehensive descriptions of the submitted algorithms linked to their rank as well as the underlying participation strategies revealed common characteristics of winning solutions. These typically include the use of multi-task learning (63%) and/or multi-stage pipelines (61%), and a focus on augmentation (100%), image preprocessing (97%), data curation (79%), and postprocessing (66%). The "typical" lead of a winning team is a computer scientist with a doctoral degree, five years of experience in biomedical image analysis, and four years of experience in deep learning. Two core general development strategies stood out for highly-ranked teams: the reflection of the metrics in the method design and the focus on analyzing and handling failure cases. According to the organizers, 43% of the winning algorithms exceeded the state of the art but only 11% completely solved the respective domain problem. The insights of our study could help researchers (1) improve algorithm development strategies when approaching new problems, and (2) focus on open research questions revealed by this work. © 2023, CC BY.

关键词： Benchmarking

Generative Artificial Intelligence in Healthcare: Ethical Considerations and Assessment Checklist

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Ning, Yilin Teixayavong, Salinelat Shang, Yuqing Savulescu, Julian Nagaraj, Vaishaanth Miao, Di Mertens, Mayli Wei Ting, Daniel Shu Ling Ong, Jasmine Chiat Liu, Mingxuan Cao, Jiuwen Dunn, Michael Vaughan, Roger Hock Ong, Marcus Eng Sung, Joseph Jao-Yiu Topol, Eric J. Liu, Nan Centre for Quantitative Medicine Duke-NUS Medical School Singapore Singapore Centre for Biomedical Ethics National University of Singapore Singapore Singapore Wellcome Centre for Ethics and Humanities University of Oxford Oxford United Kingdom School of Medicine Imperial College London London United Kingdom Centre for Ethics Department of Philosophy University of Antwerp Antwerp Belgium Antwerp Center on Responsible AI University of Antwerp Antwerp Belgium Singapore Eye Research Institute Singapore National Eye Centre Singapore Singapore SingHealth AI Office Singapore Health Services Singapore Singapore Division of Pharmacy Singapore General Hospital Singapore Singapore Machine Learning and I-Health International Cooperation Base of Zhejiang Province Hangzhou Dianzi University Zhejiang China Artificial Intelligence Institute Hangzhou Dianzi University Zhejiang China Programme in Health Services and Systems Research Duke-NUS Medical School Singapore Singapore Department of Emergency Medicine Singapore General Hospital Singapore Singapore Lee Kong Chian School of Medicine Nanyang Technological University Singapore Singapore State Key Laboratory of Digestive Disease The Chinese University of Hong Kong Hong Kong Scripps Research Translational Institute Scripps Research La Jolla CA United States Institute of Data Science National University of Singapore Singapore Singapore Centre for Quantitative Medicine Duke-NUS Medical School 8 College Road Singapore169857 Singapore

The widespread use of ChatGPT and other emerging technology powered by generative artificial intelligence (GenAI) has drawn much attention to potential ethical issues, especially in high-stakes applications such as healthcare, but ethical discussions are yet to translate into operationalisable solutions. Furthermore, ongoing ethical discussions often neglect other types of GenAI that have been used to synthesise data (e.g., images) for research and practical purposes, which resolved some ethical issues and exposed others. We conduct a scoping review of ethical discussions on GenAI in healthcare to comprehensively analyse gaps in the current research, and further propose to reduce the gaps by developing a checklist for comprehensive assessment and transparent documentation of ethical discussions in GenAI research. The checklist can be readily integrated into the current peer review and publication system to enhance GenAI research, and may be used for ethics-related disclosures for GenAI-powered products, healthcare applications of such products and beyond. © 2023, CC BY.

关键词： Artificial intelligence

167-PFlops deep learning for electron microscopy: From learning physics to atomic manipulation

学校读者我要写书评

暂无评论

167-PFlops deep learning for electron microscopy: From learn...

2018 International Conference for High Performance Computing, Networking, Storage, and Analysis, SC 2018

作者： Patton, Robert M. Travis Johnston, J. Young, Steven R. Schuman, Catherine D. March, Don D. Potok, Thomas E. Rose, Derek C. Lim, Seung-Hwan Karnowski, Thomas P. Ziatdinov, Maxim A. Kalinin, Sergei V. Oak Ridge National Laboratory Oak RidgeTN37831-6085 United States Computational Data Analytics Group United States Geographic Information Science and Technology Group United States Imaging Signals and Machine Learning Group United States Institute for Functional Imaging of Materials United States Center for Nanophase Materials Sciences United States

ISBN: (纸本)9781538683842

An artificial intelligence system called MENNDL, which used 25,200 NVIDIA Volta GPUs on Oak Ridge National Laboratory's Summit machine, automatically designed an optimal deep learning network in order to extract structural information from raw atomic-resolution microscopy data. In a few hours, MENNDL creates and evaluates millions of networks using a scalable, parallel, asynchronous genetic algorithm augmented with a support vector machine to automatically find a superior deep learning network topology and hyper-parameter set than a human expert can find in months. For the application of electron microscopy, the system furthers the goal of improving our understanding of the electron-beam-matter interactions and real-time image-based feedback, which enables a huge step beyond human capacity towards nanofabricating materials automatically. MENNDL has been scaled to the 4,200 available nodes of Summit achieving a measured 152.5 PFlops, with an estimated sustained performance of 167 PFlops when the entire machine is available. © 2018 IEEE.

关键词： Genetic algorithms

Feature extraction for hyperspectral imagery: The evolution from shallow to deep

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Rasti, Behnood Hong, Danfeng Hang, Renlong Ghamisi, Pedram Kang, Xudong Chanussot, Jocelyn Benediktsson, Jon Atli Machine Learning Group Exploration Division Helmholtz Institute Freiberg for Resource Technology Helmholtz-Zentrum Dresden-Rossendorf Freiberg09599 Germany Univ. Grenoble Alpes CNRS Grenoble INP GIPSAlab Grenoble38000 France Jiangsu Key Laboratory of Big Data Analysis Technology School of Automation Nanjing University of Information Science and Technology Nanjing210044 China Machine Learning Group Exploration Division Helmholtz Institute Freiberg for Resource Technology Helmholtz-Zentrum Dresden-Rossendorf Freiberg09599 Germany College of Electrical and Information Engineering Hunan University Changsha410082 China Key Laboratory of Visual Perception and Artificial Intelligence of Hunan Province Changsha410082 China Univ. Grenoble Alpes Inria CNRS Grenoble INP LJK GrenobleF-38000 France Faculty of Electrical and Computer Engineering University of Iceland Reykjavik101 Iceland Faculty of Electrical and Computer Engineering University of Iceland Reykjavik107 Iceland

The final version of the paper can be found in IEEE Geoscience and Remote Sensing Magazine. Hyperspectral images provide detailed spectral information through hundreds of (narrow) spectral channels (also known as dimensionality or bands) with continuous spectral information that can accurately classify diverse materials of interest. The increased dimensionality of such data makes it possible to significantly improve data information content but provides a challenge to the conventional techniques (the so-called curse of dimensionality) for accurate analysis of hyperspectral images. Feature extraction, as a vibrant field of research in the hyperspectral community, evolved through decades of research to address this issue and extract informative features suitable for data representation and classification. The advances in feature extraction have been inspired by two fields of research, including the popularization of image and signal processing as well as machine (deep) learning, leading to two types of feature extraction approaches named shallow and deep techniques. This article outlines the advances in feature extraction approaches for hyperspectral imagery by providing a technical overview of the state-of-the-art techniques, providing useful entry points for researchers at different levels, including students, researchers, and senior researchers, willing to explore novel investigations on this challenging topic. In more detail, this paper provides a bird's eye view over shallow (both supervised and unsupervised) and deep feature extraction approaches specifically dedicated to the topic of hyperspectral feature extraction and its application on hyperspectral image classification. Additionally, this paper compares 15 advanced techniques with an emphasis on their methodological foundations in terms of classification accuracies. Furthermore, to push this vibrant field of research forward an impressive amount of codes and libraries is shared at: https://***/Behnoo

关键词： Feature extraction

QUBIQ: Uncertainty Quantification for biomedical Image Segmentation Challenge

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Hongwei Bran Navarro, Fernando Ezhov, Ivan Bayat, Amirhossein Das, Dhritiman Kofler, Florian Shit, Suprosanna Waldmannstetter, Diana Paetzold, Johannes C. Hu, Xiaobin Wiestler, Benedikt Zimmer, Lucas Amiranashvili, Tamaz Prabhakar, Chinmay Berger, Christoph Weidner, Jonas Alonso-Basanta, Michelle Rashid, Arif Baid, Ujjwal Adel, Wesam Alis, Deniz Baheti, Bhakti Bai, Yingbin Bhat, Ishaan Cetindag, Sabri Can Chen, Wenting Cheng, Li Dutande, Prasad Dular, Lara Elattar, Mustafa A. Feng, Ming Gao, Shengbo Huisman, Henkjan Hu, Weifeng Innani, Shubham Ji, Wei Karimi, Davood Kuijf, Hugo J. Kwak, Jin Tae Le, Hoang Long Li, Xiang Lin, Huiyan Liu, Tongliang Ma, Jun Ma, Kai Ma, Ting Oksuz, Ilkay Holland, Robbie Oliveira, Arlindo L. Pal, Jimut Bahan Pei, Xuan Qiao, Maoying Saha, Anindo Selvan, Raghavendra Shen, Linlin Silva, Joao Lourenco Spiclin, Ziga Talbar, Sanjay Wang, Dadong Wang, Wei Wang, Xiong Wang, Yin Xi, Ruiling Xu, Kele Yang, Yanwu Yergin, Mert Yu, Shuang Zeng, Lingxi Zhang, YingLin Zhao, Jiachen Zheng, Yefeng Zukovec, Martin Do, Richard Becker, Anton Simpson, Amber Konukoglu, Ender Jakab, Andras Bakas, Spyridon Joskowicz, Leo Menze, Bjoern Department of Informatics Technical University of Munich Germany Athinoula A. Martinos Center for Biomedical Imaging Massachusetts General Hospital Harvard Medical School United States Department of Quantitative Biomedicine University of Zurich Switzerland University Children’s Hospital Zurich University of Zurich Switzerland Department of Radioncology and Radiation Theraphy Klinikum rechts der Isar Technical University of Munich Germany Department of Information Technology and Electrical Engineering ETH-Zurich Switzerland Department of Radiology Memorial Sloan Kettering Cancer Center New York City United States Department of Biomedical and Molecular Sciences Queen’s University Canada TranslaTUM - Central Institute for Translational Cancer Research Technical University of Munich Germany McGovern Institute Massachusetts Institute of Technology United States Institute for Diagnostic and Interventional Radiology Unveristy Zurich Hospital Switzerland BioMedIA Imperial College London United Kingdom Department of Radiation Oncology University of Pennsylvania PA United States University of Pennsylvania PA United States Department of Radiation Oncology Winship Cancer Institute of Emory University Georgia United States Nile University Cairo Egypt Department of Medical Sciences Acibadem University Istanbul Turkey Shri Guru Gobind Singhji Institute of Engineering and Technology Maharashtra Nanded India Trustworthy Machine Learning Lab University of Sydney Australia Image Sciences Institute University Medical Center Utrecht Netherlands Computer Engineering Department Istanbul Technical University Istanbul Turkey School of Computer Science Shenzhen University Shenzhen China University of Alberta United States University of Ljubljana Faculty of Electrical Engineering Ljubljana Slovenia Tongji University Shanghai China OPPO Research Institute Shanghai China School of Biological and Medical Engineering Beihang University Beijing China Harvard Medical School Boston

Uncertainty in medical image segmentation tasks, especially inter-rater variability, arising from differences in interpretations and annotations by various experts, presents a significant challenge in achieving consistent and reliable image segmentation. This variability not only reflects the inherent complexity and subjective nature of medical image interpretation but also directly impacts the development and evaluation of automated segmentation algorithms. Accurately modeling and quantifying this variability is essential for enhancing the robustness and clinical applicability of these algorithms. We report the set-up and summarize the benchmark results of the Quantification of Uncertainties in biomedical Image Quantification Challenge (QUBIQ), which was organized in conjunction with International Conferences on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2020 and 2021. The challenge focuses on the uncertainty quantification of medical image segmentation which considers the omnipresence of inter-rater variability in imaging datasets. The large collection of images with multi-rater annotations features various modalities such as MRI and CT;various organs such as the brain, prostate, kidney, and pancreas;and different image dimensions 2D-vs-3D. A total of 24 teams submitted different solutions to the problem, combining various baseline models, Bayesian neural networks, and ensemble model techniques. The obtained results indicate the importance of the ensemble models, as well as the need for further research to develop efficient 3D methods for uncertainty quantification methods in 3D segmentation tasks. © 2024, CC BY-NC-ND.

关键词： Image segmentation

Time-uniform, nonparametric, nonasymptotic confidence sequences

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Howard, Steven R. Ramdas, Aaditya McAuliffe, Jon Sekhon, Jasjeet Departments of Statistics Political Science UC Berkeley Departments of Statistics and Data Science United States Machine Learning Carnegie Mellon The Voleon Group United States

A confidence sequence is a sequence of confidence intervals that is uniformly valid over an unbounded time horizon. Our work develops confidence sequences whose widths go to zero, with nonasymptotic coverage guarantees under nonparametric conditions. We draw connections between the Cramér-Chernoff method for exponential concentration, the law of the iterated logarithm (LIL), and the sequential probability ratio test—our confidence sequences are time-uniform extensions of the first;provide tight, nonasymptotic characterizations of the second;and generalize the third to nonparametric settings, including sub-Gaussian and Bernstein conditions, self-normalized processes, and matrix martingales. We illustrate the generality of our proof techniques by deriving an empirical-Bernstein bound growing at a LIL rate, as well as a novel upper LIL for the maximum eigenvalue of a sum of random matrices. Finally, we apply our methods to covariance matrix estimation and to estimation of sample average treatment effect under the Neyman-Rubin potential outcomes model. Copyright © 2018, The Authors. All rights reserved.

关键词： Eigenvalues and eigenfunctions

Weakly Supervised Object Detection with 2D and 3D Regression Neural Networks

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Dubost, Florian Adams, Hieab Yilmaz, Pinar Bortsova, Gerda Van Tulder, Gijs Ikram, M. Arfan Niessen, Wiro Vernooij, Meike De Bruijne, Marleen Departments of Radiology and Medical Informatics Erasmus Medical Center Biomedical Imaging Group Rotterdam Rotterdam3015 GE Netherlands Departments of Radiology and Epidemiology Erasmus Medical Center Rotterdam3015 GE Netherlands Faculty of Applied Science Department of Imaging Physics Tu Delft Netherlands Department of Computer Science Machine Learning Section University of Copenhagen CopenhagenDK-2110 Denmark

Weakly supervised detection methods can infer the location of target objects in an image without requiring location or appearance information during training. We propose a weakly supervised deep learning method for the detection of objects that appear at multiple locations in an image. The method computes attention maps using the last feature maps of an encoder-decoder network optimized only with global labels: The number of occurrences of the target object in an image. In contrast with previous approaches, attention maps are generated at full input resolution thanks to the decoder part. The proposed approach is compared to multiple state-of-the-art methods in two tasks: The detection of digits in MNIST-based datasets, and the real life application of detection of enlarged perivascular spaces- A type of brain lesion-in four brain regions in a dataset of 2202 3D brain MRI scans. In MNIST-based datasets, the proposed method outperforms the other methods. In the brain dataset, several weakly supervised detection methods come close to the human intrarater agreement in each region. The proposed method reaches the lowest number of false positive detections in all brain regions at the operating point, while its average sensitivity is similar to that of the other best methods. Copyright © 2019, The Authors. All rights reserved.

关键词： Location

AIROGS: ARTIFICIAL INTELLIGENCE FOR ROBUST GLAUCOMA SCREENING CHALLENGE

学校读者我要写书评

暂无评论

arXiv 2023年

作者： De Vente, Coen Vermeer, Koenraad A. Jaccard, Nicolas Wang, He Sun, Hongyi Khader, Firas Truhn, Daniel Aimyshev, Temirgali Zhanibekuly, Yerkebulan Le, Tien-Dung Galdran, Adrian Ballester, Miguel Ángel González Carneiro, Gustavo Devika, R.G. Hrishikesh, P.S. Puthussery, Densen Liu, Hong Yang, Zekang Kondo, Satoshi Kasai, Satoshi Wang, Edward Durvasula, Ashritha Heras, Jónathan Zapata, Miguel Ángel Araújo, Teresa Aresta, Guilherme Bogunović, Hrvoje Arikan, Mustafa Lee, Yeong Chan Cho, Hyun Bin Choi, Yoon Ho Qayyum, Abdul Razzak, Imran Van Ginneken, Bram Lemij, Hans G. Sánchez, Clara I. Group Informatics Institute Universiteit Van Amsterdam Noord-Holland Amsterdam Netherlands The Department of Biomedical Engineering and Physics Amsterdam Umc Locatie Amc Noord-Holland Amsterdam Netherlands Department of Radiology and Nuclear Medicine Radboudumc Gelderland Nijmegen Netherlands The Rotterdam Ophthalmic Institute Rotterdam Eye Hospital Rotterdam Netherlands Project Orbis International Inc. New York United States The Peking Union Medical College Hospital Beijing China The Xuanwu Hospital Capital Medical University Beijing100053 China The Tsinghua University Beijing China The Department of Diagnostic and Interventional Radiology University Hospital Aachen Aachen Germany Cmc Technologies Llp Nur-Sultan Kazakhstan Kbc Belgium Universitat Pompeu Fabra Barcelona Spain The Australian Institute for Machine Learning Aiml University of Adelaide Australia Icrea Barcelona Spain The Centre for Vision Speech and Signal Processing University of Surrey United Kingdom The College of Engineering Trivandrum India Founding Minds Software India The Institute of Computing Technology Chinese Academy of Sciences China The Muroran Institute of Technology Japan The Niigata University of Health and Welfare Japan The Schulich School of Medicine and Dentistry University of Western Ontario London Canada The Department of Mathematics and Computer Science University of la Rioja Spain Hospital Vall Hebron Passeig Roser 126 Sant Cugat del Vallès Barcelona08195 Spain UPRetina Barcelona Spain The Christian Doppler Laboratory for Artificial Intelligence in Retina Department of Ophthalmology and Optometry Medical University of Vienna Vienna Austria The Institute of Ophthalmology University College London London United Kingdom The Research Institute for Future Medicine Samsung Medical Center Seoul Korea Republic of Sungkyunkwan University Samsung Medical Center Seoul Korea Republic of The Department of Artificial Intelligence and Informatics Mayo Clinic Jacks

The early detection of glaucoma is essential in preventing visual impairment. Artificial intelligence (AI) can be used to analyze color fundus photographs (CFPs) in a cost-effective manner, making glaucoma screening more accessible. While AI models for glaucoma screening from CFPs have shown promising results in laboratory settings, their performance decreases significantly in realworld scenarios due to the presence of out-of-distribution and low-quality images. To address this issue, we propose the Artificial Intelligence for Robust Glaucoma Screening (AIROGS) challenge. This challenge includes a large dataset of around 113,000 images from about 60,000 patients and 500 different screening centers, and encourages the development of algorithms that are robust to ungradable and unexpected input data. We evaluated solutions from 14 teams in this paper, and found that the best teams performed similarly to a set of 20 expert ophthalmologists and optometrists. The highest-scoring team achieved an area under the receiver operating characteristic curve of 0.99 (95% CI: 0.98-0.99) for detecting ungradable images on-the-fly. Additionally, many of the algorithms showed robust performance when tested on three other publicly available datasets. These results demonstrate the feasibility of robust AI-enabled glaucoma screening. Copyright © 2023, The Authors. All rights reserved.

关键词： Diagnosis