A5063535548- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Neural Networks and Applications
- Gaussian Processes and Bayesian Inference
- COVID-19 diagnosis using AI
- Genomics and Phylogenetic Studies
- Functional Brain Connectivity Studies
- EEG and Brain-Computer Interfaces
- COVID-19 Clinical Research Studies
- Machine Learning in Bioinformatics
- Machine Learning and Data Classification
- ECG Monitoring and Analysis
- RNA and protein synthesis mechanisms
- Advanced Image and Video Retrieval Techniques
- Sepsis Diagnosis and Treatment
- Distributed and Parallel Computing Systems
- COVID-19 and healthcare impacts
- Machine Learning and Algorithms
- Image Processing and 3D Reconstruction
- Aesthetic Perception and Analysis
- Digital Media Forensic Detection
- Medical Image Segmentation Techniques
- Radiomics and Machine Learning in Medical Imaging
- Domain Adaptation and Few-Shot Learning
- Model Reduction and Neural Networks
Technical University of Denmark
2018-2023
Oceaneering International (United States)
2023
Stanford University
2023
NED University of Engineering and Technology
2023
National Grid (United States)
2022
University of Copenhagen
2020
How we choose to represent our data has a fundamental impact on ability subsequently extract information from them. Machine learning promises automatically determine efficient representations large unstructured datasets, such as those arising in biology. However, empirical evidence suggests that seemingly minor changes these machine models yield drastically different result biological interpretations of data. This begs the question what even constitutes most meaningful representation. Here,...
A main problem with reproducing machine learning publications is the variance of metric implementations across papers.A lack standardization leads to different behavior in mechanisms such as checkpointing, rate schedulers or early stopping, that will influence reported results.For example, a complex Fréchet inception distance (FID) for synthetic image quality evaluation (Heusel et al., 2017) differ based on specific interpolation method used.
Abstract Patients with severe COVID-19 have overwhelmed healthcare systems worldwide. We hypothesized that machine learning (ML) models could be used to predict risks at different stages of management and thereby provide insights into drivers prognostic markers disease progression death. From a cohort approx. 2.6 million citizens in Denmark, SARS-CoV-2 PCR tests were performed on subjects suspected for disease; 3944 cases had least one positive test subjected further analysis. from the...
Pulmonary opacification is the inflammation in lungs caused by many respiratory ailments, including novel corona virus disease 2019 (COVID-19). Chest X-rays (CXRs) with such opacifications render regions of imperceptible, making it difficult to perform automated image analysis on them. In this work, we focus segmenting from abnormal CXRs as part a pipeline aimed at risk scoring COVID-19 CXRs. We treat high opacity missing data and present modified CNN-based segmentation network that utilizes...
Spatial Transformer layers allow neural networks, at least in principle, to be invariant large spatial transformations image data. The model has, however, seen limited uptake as most practical implementations support only that are too restricted, e.g. affine or homographic maps, and/or destructive such thin plate splines. We investigate the use of flexible diffeomorphic within networks and demonstrate significant performance gains can attained over currently-used models. learned found both...
We propose and investigate new complementary methodologies for estimating predictive variance networks in regression neural networks. derive a locally aware mini-batching scheme that result sparse robust gradients, show how to make unbiased weight updates network. Further, we formulate heuristic robustly fitting both the mean post hoc. Finally, take inspiration from posterior Gaussian processes network architecture with similar extrapolation properties processes. The proposed are...
Disentangled representation learning finds compact, independent and easy-to-interpret factors of the data. Learning such has been shown to require an inductive bias, which we explicitly encode in a generative model images. Specifically, propose with two latent spaces: one that represents spatial transformations input data, another transformed We find latter naturally captures intrinsic appearance To realize model, Variationally Inferred Transformational Autoencoder (VITAE) incorporates...
ABSTRACT Background Patients with severe COVID-19 have overwhelmed healthcare systems worldwide. We hypothesized that Machine Learning (ML) models could be used to predict risks at different stages of management (at diagnosis, hospital admission and ICU admission) thereby provide insights into drivers prognostic markers disease progression death. Methods From a cohort approx. 2.6 million citizens in the two regions Denmark, SARS-CoV-2 PCR tests were performed on subjects suspected for...
<p>Major depressive disorder (MDD) is a common mental affecting the lives of about 280 million people and increasing rates suicidal mortality. The current methods diagnosis depression are subjective, time-consuming, expensive, inaccurate because its heterogeneous symptoms that overlap with other disorders. In this paper, we exploit potential fusion artificial intelligence (AI) electroencephalogram (EEG) to revolutionize automatic compare classification performance machine learning (ML)...
<p>Major depressive disorder (MDD) is a common mental affecting the lives of about 280 million people and increasing rates suicidal mortality. The current methods diagnosis depression are subjective, time-consuming, expensive, inaccurate because its heterogeneous symptoms that overlap with other disorders. In this paper, we exploit potential fusion artificial intelligence (AI) electroencephalogram (EEG) to revolutionize automatic compare classification performance machine learning (ML)...
Established methods for unsupervised representation learning such as variational autoencoders produce none or poorly calibrated uncertainty estimates making it difficult to evaluate if learned representations are stable and reliable. In this work, we present a Bayesian autoencoder learning, which is trained using novel lower-bound of the evidence. This maximized Monte Carlo EM with distribution that takes shape Laplace approximation. We develop new Hessian approximation scales linearly data...
The encoder network of an autoencoder is approximation the nearest point projection onto manifold spanned by decoder. A concern with this that, while output always unique, can possibly have infinitely many values. This implies that latent representations learned be misleading. Borrowing from geometric measure theory, we introduce idea using reach decoder to determine if optimal exists for a given dataset and We develop local generalization propose numerical estimator thereof. demonstrate...