Publications
Journal Article
Artificial intelligence in dry eye disease
The Ocular Surface 23 (2022): 74-86.Status: Published
Artificial intelligence in dry eye disease
Dry eye disease (DED) has a prevalence of between 5 and 50%, depending on the diagnostic criteria used and population under study. However, it remains one of the most underdiagnosed and undertreated conditions in ophthalmology. Many tests used in the diagnosis of DED rely on an experienced observer for image interpretation, which may be considered subjective and result in variation in diagnosis. Since artificial intelligence (AI) systems are capable of advanced problem solving, use of such techniques could lead to more objective diagnosis. Although the term ‘AI’ is commonly used, recent success in its applications to medicine is mainly due to advancements in the sub-field of machine learning, which has been used to automatically classify images and predict medical outcomes. Powerful machine learning techniques have been harnessed to understand nuances in patient data and medical images, aiming for consistent diagnosis and stratification of disease severity. This is the first literature review on the use of AI in DED. We provide a brief introduction to AI, report its current use in DED research and its potential for application in the clinic. Our review found that AI has been employed in a wide range of DED clinical tests and research applications, primarily for interpretation of interferometry, slit-lamp and meibography images. While initial results are promising, much work is still needed on model development, clinical testing and standardisation.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2022 |
Journal | The Ocular Surface |
Volume | 23 |
Pagination | 74 - 86 |
Date Published | Jan-01-2022 |
Publisher | Elsevier |
ISSN | 15420124 |
Keywords | artificial intelligence, Dry eye disease, Machine learning |
URL | https://linkinghub.elsevier.com/retrieve/pii/S1542012421001324 |
DOI | 10.1016/j.jtos.2021.11.004 |
Bias og kvantitativ analyse innen velferd
Tidsskrift for velferdsforskning 25, no. 3 (2022): 1-24.Status: Published
Bias og kvantitativ analyse innen velferd
Ifølge Norges nasjonale strategi for kunstig intelligens (2020) er offentlig forvaltning og helse blant Norges satsningsområder for bruk av kunstig intelligens. Maskinlæring er en undergruppe av kunstig intelligens med potensiale for å løse en rekke utfordringer, men som også gir opphav til utfordringer. En slik utfordring er bias, eller skjevhet. Et eksempel på skjevhet er at tilstedeværende ulikheter i samfunnet representeres i datagrunnlaget maskinlæringsmodeller utvikles på. De resulterende modellene står dermed i fare for å adoptere og videreføre disse ulikhetene. En utfordring er at skjevhet har ulike definisjoner innen ulike fagområder, og kan ha mange ulike opphav. Vi bidrar til å løse denne utfordringen ved å gi en oversikt over ulike typer skjevhet og deres opphav med illustrasjoner fra et velferdsperspektiv, samt avklarer forskjellen til det nærliggende konseptet rettferdighet. Vi demonstrerer utfordringer relatert til databaserte modellers oppførsel ved å benytte maskinlæring til å predikere fremtidig ressursbehov i helsevesenet, spesifikt antall legebesøk i kommuner. Vi demonstrerer ulike typer skjevheter, diskuterer mulige løsninger og bruker metoder fra forklarbar kunstig intelligens for å analysere opphavet til skjevheter i forklaringsvariablene. Det finnes ingen universell løsning for å håndtere alle typer skjevheter, men skjevhet må tas høyde for i alle deler av en kvantitativ analyse.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2022 |
Journal | Tidsskrift for velferdsforskning |
Volume | 25 |
Issue | 3 |
Pagination | 1-24 |
Publisher | Universitetsforlaget |
Place Published | Tidsskrift for velferdsforskning |
Keywords | forklarbar kunstig intelligens, Maskinlæring, skjevhetsbegreper, velferdsforskning, XGBoost |
DOI | 10.18261/tfv.25.3.3 |
On evaluation metrics for medical applications of artificial intelligence
Scientific Reports 12 (2022): 1-9.Status: Published
On evaluation metrics for medical applications of artificial intelligence
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2022 |
Journal | Scientific Reports |
Volume | 12 |
Number | 1 |
Pagination | 1–9 |
Publisher | Nature Publishing Group |
To explain or not to explain?—Artificial intelligence explainability in clinical decision support systems
PLOS Digital Health 1, no. 2 (2022): e0000016.Status: Published
To explain or not to explain?—Artificial intelligence explainability in clinical decision support systems
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2022 |
Journal | PLOS Digital Health |
Volume | 1 |
Issue | 2 |
Pagination | e0000016 |
Date Published | May-02-2023 |
Publisher | PLOS |
URL | https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.... |
DOI | 10.1371/journal.pdig.000001610.1371/ |
Public outreach
Data samlet fra åpne kilder er ikke ufarlig
Kronikk: Aftenposten, 2022.Status: Published
Data samlet fra åpne kilder er ikke ufarlig
Afilliation | Communication Systems |
Project(s) | The Center for Resilient Networks and Applications, Department of Holistic Systems |
Publication Type | Public outreach |
Year of Publication | 2022 |
Publisher | Aftenposten |
Place Published | Kronikk |
Proceedings, refereed
Experiences and Lessons Learned from a Crowdsourced-Remote Hybrid User Survey Framework
In 2022 IEEE International Symposium on Multimedia (ISM). Italy: IEEE, 2022.Status: Published
Experiences and Lessons Learned from a Crowdsourced-Remote Hybrid User Survey Framework
Subjective user studies are important to ensure the fidelity and usability of systems that generate multimedia content. Testing how end-users and domain experts perceive multimedia assets might provide crucial information. In this paper, we present our experiences with the open source hybrid crowdsourced-remote user survey framework called Huldra, which is intended for conducting web-based subjective user studies and aims to integrate the individual benefits associated with traditional, crowdsourced, and remote methods. We disseminate our experiences and insights from two actively deployed use cases and discuss challenges and opportunities associated with using Huldra as a framework for conducting user studies.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2022 |
Conference Name | 2022 IEEE International Symposium on Multimedia (ISM) |
Pagination | 161-162 |
Publisher | IEEE |
Place Published | Italy |
URL | https://ieeexplore.ieee.org/document/10019678 |
DOI | 10.1109/ISM55400.2022.00035 |
Explainability methods for machine learning systems for multimodal medical datasets: research proposal
In ACM Multimedia Systems (MMSys) Conference. ACM, 2022.Status: Published
Explainability methods for machine learning systems for multimodal medical datasets: research proposal
This paper contains the research proposal of Andrea M. Storås that was presented at the MMSys 2022 doctoral symposium. Machine learning models have the ability to solve medical tasks with a high level of performance, e.g., classifying medical videos and detecting anomalies using different sources of data. However, many of these models are highly complex and difficult to understand. Lack of interpretability can limit the use of machine learning systems in the medical domain. Explainable artificial intelligence provides explanations regarding the models and their predictions. In this PhD project, we develop machine learning models for automatic analysis of medical data and explain the results using established techniques from the field of explainable artificial intelligence. Current research indicate that there are still open issues to be solved in order for end users to understand multimedia systems powered by machine learning. Consequently, new explanation techniques will also be developed. Different types of medical data are applied in order to investigate the generalizability of the methods.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2022 |
Conference Name | ACM Multimedia Systems (MMSys) Conference |
Pagination | 347-351 |
Publisher | ACM |
ISBN Number | 978-1-4503-9283-9/22/06 |
DOI | 10.1145/3524273.3533925 |
Huldra: A Framework for Collecting Crowdsourced Feedback on Multimedia Assets
In Proceedings of the 13th ACM Multimedia Systems Conference (MMSys '22). ACM, 2022.Status: Published
Huldra: A Framework for Collecting Crowdsourced Feedback on Multimedia Assets
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2022 |
Conference Name | Proceedings of the 13th ACM Multimedia Systems Conference (MMSys '22) |
Pagination | 203-209 |
Publisher | ACM |
DOI | 10.1145/3524273.3532887 |
Predicting Tacrolimus Exposure in Kidney Transplanted Patients Using Machine Learning
In 35th IEEE CBMS International Symposium on Computer-Based Medical Systems. IEEE, 2022.Status: Published
Predicting Tacrolimus Exposure in Kidney Transplanted Patients Using Machine Learning
Tacrolimus is one of the cornerstone immunosuppressive drugs in most transplantation centers worldwide following solid organ transplantation. Therapeutic drug monitoring of tacrolimus is necessary in order to avoid rejection of the transplanted organ or severe side effects. However, finding the right dose for a given patient is challenging, even for experienced clinicians. Consequently, a tool that can accurately estimate the drug exposure for individual dose adaptions would be of high clinical value. In this work, we propose a new technique using machine learning to estimate the tacrolimus exposure in kidney transplant recipients. Our models achieve predictive errors that are at the same level as an established population pharmacokinetic model, but are faster to develop and require less knowledge about the pharmacokinetic properties of the drug.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2022 |
Conference Name | 35th IEEE CBMS International Symposium on Computer-Based Medical Systems |
Pagination | 38-43 |
Publisher | IEEE |
Keywords | Machine learning, personalized medicine, transplantation |
DOI | 10.1109/CBMS55023.2022.00014 |
Talks, invited
Explainable Artificial Intelligence in Medicine
In Nordic AI Meet 2022, 2022.Status: Accepted
Explainable Artificial Intelligence in Medicine
Machine learning (ML) has shown outstanding abilities to solve a large variety of tasks such as image recognition and natural language processing, which has huge relevance for the medical field. Complex ML models, including convolutional neural networks (CNNs), are used to analyse high dimensional data such as images and videos from medical examinations. With increasing model complexity, the demand for techniques improving human understanding of the ML models also increases. If medical doctors do not understand how the models work, they might not know when the models are actually wrong or even refuse to use them. This can hamper the implementation of ML systems in the clinic and negatively affect patients. To promote successful integration of ML systems in the clinic, it is important to provide explanations that establish trust in the models among healthcare personnel. Explainable artificial intelligence (XAI) aims to provide explanations about ML models and their predictions. Several techniques have already been developed. Existing XAI methods often fail to meet the requirements of medical doctors, probably because they are not sufficiently involved in the development of the methods. We develop ML models solving tasks in various medical domains. The resulting models are explained using a selection of existing XAI methods, and the explanations are evaluated by medical experts. Their feedback is used to develop improved XAI methods. We have investigated established techniques for making ML systems more transparent in the fields of gastroenterology, assisted reproductive technology, organ transplantation and cardiology. Experiences from our projects will be used to develop new explanation techniques for clinical practice in close collaboration with medical experts.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Talks, invited |
Year of Publication | 2022 |
Location of Talk | Nordic AI Meet 2022 |
Keywords | Explainable artificial intelligence, Machine learning, medicine |
Poster
Predicting drug exposure in kidney transplanted patients using machine learning
NORA Annual Conference, Stavanger, Norway, 2022.Status: Published
Predicting drug exposure in kidney transplanted patients using machine learning
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Poster |
Year of Publication | 2022 |
Place Published | NORA Annual Conference, Stavanger, Norway |
Type of Work | Poster presentation |
Miscellaneous
Visual explanations for polyp detection: How medical doctors assess intrinsic versus extrinsic explanations
arXiv, 2022.Status: Published
Visual explanations for polyp detection: How medical doctors assess intrinsic versus extrinsic explanations
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Miscellaneous |
Year of Publication | 2022 |
Publisher | arXiv |
URL | https://arxiv.org/abs/2204.00617 |
DOI | 10.48550/arXiv.2204.00617 |
Book Chapter
Artificial Intelligence in Gastroenterology
In Artificial Intelligence in Medicine, 1-20. Cham: Springer International Publishing, 2021.Status: Published
Artificial Intelligence in Gastroenterology
The holy grail in endoscopy examinations has for a long time been assisted diagnosis using Artificial Intelligence (AI). Recent developments in computer hardware are now enabling technology to equip clinicians with promising tools for computer-assisted diagnosis (CAD) systems. However, creating viable models or architectures, training them, and assessing their ability to diagnose at a human level, are complicated tasks. This is currently an active area of research, and many promising methods have been proposed. In this chapter, we give an overview of the topic. This includes a description of current medical challenges followed by a description of the most commonly used methods in the field. We also present example results from research targeting some of these challenges, and a discussion on open issues and ongoing work is provided. Hopefully, this will inspire and enable readers to future develop CAD systems for gastroenterology.
Afilliation | Communication Systems, Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Book Chapter |
Year of Publication | 2021 |
Book Title | Artificial Intelligence in Medicine |
Pagination | 1 - 20 |
Date Published | 09/2021 |
Publisher | Springer International Publishing |
Place Published | Cham |
ISBN Number | 978-3-030-58080-3 |
Keywords | Anomaly detection, artificial intelligence, Gastrointestinal endoscopy, Hand-crafted features, Neural Networks, Performance, Semantic segmentation |
URL | https://link.springer.com/referenceworkentry/10.1007%2F978-3-030-58080-3... |
DOI | 10.1007/978-3-030-58080-3_163-2 |
Proceedings, refereed
Data Augmentation Using Generative Adversarial Networks For Creating Realistic Artificial Colon Polyp Images
In DDW 2021, 2021.Status: Published
Data Augmentation Using Generative Adversarial Networks For Creating Realistic Artificial Colon Polyp Images
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2021 |
Conference Name | DDW 2021 |
Impact of Image Resolution on Convolutional Neural Networks Performance in Gastrointestinal Endoscopy
In DDW 2021, 2021.Status: Published
Impact of Image Resolution on Convolutional Neural Networks Performance in Gastrointestinal Endoscopy
Convolutional neural networks (CNNs) are increasingly used to improve and automate processes in gastroenterology, like the detection of polyps during a colonoscopy. An important input to these methods is images and videos. Up until now, no well-defined, common understanding or standard regarding the resolution of the images and video frames has been defined, and to reduce processing time and resource requirements, images are today almost always down-sampled. However, how such down-sampling and the image resolution influence the performance in context with medical data is unknown. In this work, we investigate how the resolution relates to the performance of convolutional neural networks. This can help set standards for image or video characteristics for future CNN based models in gastrointestinal endoscopy.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Proceedings, refereed |
Year of Publication | 2021 |
Conference Name | DDW 2021 |
Journal Article
DeepFake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine
Nature Scientific Reports 11 (2021): 21896.Status: Published
DeepFake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine
Recent global developments underscore the prominent role big data have in modern medical science. But privacy issues constitute a prevalent problem for collecting and sharing data between researchers. However, synthetic data generated to represent real data carrying similar information and distribution may alleviate the privacy issue. In this study, we present generative adversarial networks (GANs) capable of generating realistic synthetic DeepFake 10-s 12-lead electrocardiograms (ECGs). We have developed and compared two methods, named WaveGAN* and Pulse2Pulse. We trained the GANs with 7,233 real normal ECGs to produce 121,977 DeepFake normal ECGs. By verifying the ECGs using a commercial ECG interpretation program (MUSE 12SL, GE Healthcare), we demonstrate that the Pulse2Pulse GAN was superior to the WaveGAN* to produce realistic ECGs. ECG intervals and amplitudes were similar between the DeepFake and real ECGs. Although these synthetic ECGs mimic the dataset used for creation, the ECGs are not linked to any individuals and may thus be used freely. The synthetic dataset will be available as open access for researchers at OSF.io and the DeepFake generator available at the Python Package Index (PyPI) for generating synthetic ECGs. In conclusion, we were able to generate realistic synthetic ECGs using generative adversarial neural networks on normal ECGs from two population studies, thereby addressing the relevant privacy issues in medical datasets.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | Nature Scientific Reports |
Volume | 11 |
Pagination | 21896 |
Date Published | 09/2021 |
Publisher | Springer nature |
URL | https://www.nature.com/articles/s41598-021-01295-2 |
DOI | 10.1038/s41598-021-01295-2 |
DeepFake electrocardiograms: the key for open science for artificial intelligence in medicine
Scientific Reports (2021).Status: Accepted
DeepFake electrocardiograms: the key for open science for artificial intelligence in medicine
Recent global developments underscore the prominent role big data have in modern medical science. Privacy issues are a prevalent problem for collecting and sharing data between researchers. Synthetic data generated to represent real data carrying similar information and distribution may alleviate the privacy issue.In this study, we present generative adversarial networks (GANs) capable of generating realistic synthetic DeepFake 12-lead 10-sec electrocardiograms (ECGs). We have developed and compare two methods, WaveGAN* and Pulse2Pulse GAN. We trained the GANs with 7,233 real normal ECG to produce 121,977 DeepFake normal ECGs. By verifying the ECGs using a commercial ECG interpretation program (MUSE 12SL, GE Healthcare), we demonstrate that the Pulse2Pulse GAN was superior to the WaveGAN to produce realistic ECGs. ECG intervals and amplitudes were similar between the DeepFake and real ECGs. These synthetic ECGs are fully anonymous and cannot be referred to any individual, hence they may be used freely. The synthetic dataset will be available as open access for researchers at OSF.io and the DeepFake generator available at the Python Package Index (PyPI) for generating synthetic ECGs.In conclusion, we were able to generate realistic synthetic ECGs using adversarial neural networks on normal ECGs from two population studies, i.e., there by addressing the relevant privacy issues in medical datasets.Competing Interest StatementThe authors have declared no competing interest.Clinical TrialN/AFunding StatementThis work is funded in part by Novo Nordisk Foundation project number NNF18CC0034900.Author DeclarationsI confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.YesThe details of the IRB/oversight body that provided approval or exemption for the research described are given below:We confirm that all experiments were performed in accordance with Helsinki guidelines and regulations of the Danish Regional Committees for Medical and Health Research Ethics. The data studies were approved by the ethical committee of Region Zealand (SJ-113, SJ-114, SJ-191), ethical committee of Copenhagen Amt (KA 98 155).All necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived.YesI understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).YesI have followed all appropriate research reporting guidelines and uploaded the relevant EQUATOR Network research reporting checklist(s) and other pertinent material as supplementary files, if applicable.YesThe Normal DeepFake ECGs are available at OSF (https://osf.io/6hved/) with corresponding MUSE 12SL ground truth values freely downloadable and usable for ECG algorithm development. The DeepFake generative model is available at https://pypi.org/project/deepfake-ecg/ to generate only synthetic ECGs. https://osf.io/6hved/ https://pypi.org/project/deepfake-ecg/
Afilliation | Scientific Computing, Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | Scientific Reports |
Publisher | Nature Publishing Group |
URL | https://www.medrxiv.org/content/early/2021/05/10/2021.04.27.21256189.1 |
DOI | 10.1101/2021.04.27.21256189 |
Explaining deep neural networks for knowledge discovery in electrocardiogram analysis
Scientific Reports 11 (2021): 10949.Status: Published
Explaining deep neural networks for knowledge discovery in electrocardiogram analysis
Deep learning-based tools may annotate and interpret medical data more quickly, consistently, and accurately than medical doctors. However, as medical doctors are ultimately responsible for clinical decision-making, any deep learning-based prediction should be accompanied by an explanation that a human can understand. We present an approach called electrocardiogram gradient class activation map (ECGradCAM), which is used to generate attention maps and explain the reasoning behind deep learning-based decision-making in ECG analysis. Attention maps may be used in the clinic to aid diagnosis, discover new medical knowledge, and identify novel features and characteristics of medical tests. In this paper, we showcase how ECGradCAM attention maps can unmask how a novel deep learning model measures both amplitudes and intervals in 12-lead electrocardiograms, and we show an example of how attention maps may be used to develop novel ECG features.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | Scientific Reports |
Volume | 11 |
Pagination | 10949 |
Date Published | 05/2021 |
Publisher | Springer Nature |
URL | http://www.nature.com/articles/s41598-021-90285-5h |
DOI | 10.1038/s41598-021-90285-5 |
Impact of Image Resolution on Deep Learning Performance in Endoscopy Image Classification: An Experimental Study Using a Large Dataset of Endoscopic Images
Diagnostics 11, no. 12 (2021).Status: Published
Impact of Image Resolution on Deep Learning Performance in Endoscopy Image Classification: An Experimental Study Using a Large Dataset of Endoscopic Images
Recent trials have evaluated the efficacy of deep convolutional neural network (CNN)-based AI systems to improve lesion detection and characterization in endoscopy. Impressive results are achieved, but many medical studies use a very small image resolution to save computing resources at the cost of losing details. Today, no conventions between resolution and performance exist, and monitoring the performance of various CNN architectures as a function of image resolution provides insights into how subtleties of different lesions on endoscopy affect performance. This can help set standards for image or video characteristics for future CNN-based models in gastrointestinal (GI) endoscopy. This study examines the performance of CNNs on the HyperKvasir dataset, consisting of 10,662 images from 23 different findings. We evaluate two CNN models for endoscopic image classification under quality distortions with image resolutions ranging from 32 × 32 to 512 × 512 pixels. The performance is evaluated using two-fold cross-validation and F1-score, maximum Matthews correlation coefficient (MCC), precision, and sensitivity as metrics. Increased performance was observed with higher image resolution for all findings in the dataset. MCC was achieved at image resolutions between 512 × 512 pixels for classification for the entire dataset after including all subclasses. The highest performance was observed with an MCC value of 0.9002 when the models were trained on the highest resolution and tested on the same resolution. Different resolutions and their effect on CNNs are explored. We show that image resolution has a clear influence on the performance which calls for standards in the field in the future.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | Diagnostics |
Volume | 11 |
Issue | 12 |
Date Published | 09/2021 |
Publisher | MDPI |
URL | https://www.mdpi.com/2075-4418/11/12/2183 |
DOI | 10.3390/diagnostics11122183 |
Model independent feature attributions: Shapley values that uncover non-linear dependencies
PeerJ Computer Science 7 (2021): e582.Status: Published
Model independent feature attributions: Shapley values that uncover non-linear dependencies
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | PeerJ Computer Science |
Volume | 7 |
Pagination | e582 |
Date Published | Jan-01-2021 |
Publisher | PeerJ |
URL | https://peerj.com/articles/cs-582 |
DOI | 10.7717/peerj-cs.582 |
Shapley Values for Feature Selection: The Good, the Bad, and the Axioms
IEEE Access 9 (2021): 144352-144360.Status: Published
Shapley Values for Feature Selection: The Good, the Bad, and the Axioms
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | IEEE Access |
Volume | 9 |
Pagination | 144352-144360 |
Publisher | IEEE |
URL | https://ieeexplore.ieee.org/document/9565902 |
DOI | 10.1109/ACCESS.2021.3119110 |
The social dilemma in artificial intelligence development and why we have to solve it
AI and Ethics (2021).Status: Published
The social dilemma in artificial intelligence development and why we have to solve it
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2021 |
Journal | AI and Ethics |
Date Published | Aug-12-2021 |
Publisher | Springer |
ISSN | 2730-5953 |
URL | https://link.springer.com/article/10.1007/s43681-021-00120-w |
DOI | 10.1007/s43681-021-00120-w |
Journal Article
Shapley Value Confidence Intervals for Attributing Variance Explained
Frontiers in Applied Mathematics and Statistics 6 (2020): 58.Status: Published
Shapley Value Confidence Intervals for Attributing Variance Explained
The coefficient of determination, the R2, is often used to measure the variance explained by an affine combination of multiple explanatory covariates. An attribution of this explanatory contribution to each of the individual covariates is often sought in order to draw inference regarding the importance of each covariate with respect to the response phenomenon. A recent method for ascertaining such an attribution is via the game theoretic Shapley value decomposition of the coefficient of determination. Such a decomposition has the desirable efficiency, monotonicity, and equal treatment properties. Under a weak assumption that the joint distribution is pseudo-elliptical, we obtain the asymptotic normality of the Shapley values. We then utilize this result in order to construct confidence intervals and hypothesis tests for Shapley values. Monte Carlo studies regarding our results are provided. We found that our asymptotic confidence intervals required less computational time to competing bootstrap methods and are able to exhibit improved coverage, especially on small samples. In an expository application to Australian real estate price modeling, we employ Shapley value confidence intervals to identify significant differences between the explanatory contributions of covariates, between models, which otherwise share approximately the same R2 value. These different models are based on real estate data from the same periods in 2019 and 2020, the latter covering the early stages of the arrival of the novel coronavirus, COVID-19.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2020 |
Journal | Frontiers in Applied Mathematics and Statistics |
Volume | 6 |
Pagination | 58 |
Date Published | Mar-12-2020 |
Publisher | Frontiers in Applied Mathematics and Statistics |
URL | https://www.frontiersin.org/articles/10.3389/fams.2020.587199/fullhttps:... |
DOI | 10.3389/fams.2020.58719910.3389/fams.2020.587199.s001 |
Shapley Value Confidence Intervals for Attributing Variance Explained
Frontiers in Applied Mathematics and Statistics 6 (2020).Status: Published
Shapley Value Confidence Intervals for Attributing Variance Explained
{The coefficient of determination, the , is often used to measure the variance explained by an affine combination of multiple explanatory covariates. An attribution of this explanatory contribution to each of the individual covariates is often sought in order to draw inference regarding the importance of each covariate with respect to the response phenomenon. A recent method for ascertaining such an attribution is via the game theoretic Shapley value decomposition of the coefficient of determination. Such a decomposition has the desirable efficiency, monotonicity, and equal treatment properties. Under a weak assumption that the joint distribution is pseudo-elliptical, we obtain the asymptotic normality of the Shapley values. We then utilize this result in order to construct confidence intervals and hypothesis tests for Shapley values. Monte Carlo studies regarding our results are provided. We found that our asymptotic confidence intervals required less computational time to competing bootstrap methods and are able to exhibit improved coverage, especially on small samples. In an expository application to Australian real estate price modeling, we employ Shapley value confidence intervals to identify significant differences between the explanatory contributions of covariates, between models, which otherwise share approximately the same value. These different models are based on real estate data from the same periods in 2019 and 2020, the latter covering the early stages of the arrival of the novel coronavirus, COVID-19.
Afilliation | Machine Learning |
Project(s) | Department of Holistic Systems |
Publication Type | Journal Article |
Year of Publication | 2020 |
Journal | Frontiers in Applied Mathematics and Statistics |
Volume | 6 |
Publisher | Frontiers |
ISSN | 2297-4687 |
URL | https://www.frontiersin.org/articles/10.3389/fams.2020.587199 |
DOI | 10.3389/fams.2020.587199 |