A5083548477- Neural Networks and Applications
- Face and Expression Recognition
- Hormonal Regulation and Hypertension
- Theoretical and Computational Physics
- nanoparticles nucleation surface interactions
- Blind Source Separation Techniques
- Image Retrieval and Classification Techniques
- Adrenal and Paraganglionic Tumors
- Machine Learning and Data Classification
- Machine Learning and Algorithms
- Gene expression and cancer classification
- Advanced Image and Video Retrieval Techniques
- Statistical Mechanics and Entropy
- Cardiac Arrhythmias and Treatments
- Cardiac pacing and defibrillation studies
- Religion, Society, and Development
- Cancer, Hypoxia, and Metabolism
- Fault Detection and Control Systems
- Machine Learning and ELM
- Data Visualization and Analytics
- Atrial Fibrillation Management and Outcomes
- Surface and Thin Film Phenomena
- Advanced Data Compression Techniques
- Remote-Sensing Image Classification
- Model Reduction and Neural Networks
University of Birmingham
2021-2025
University of Groningen
2015-2024
Creative Commons
2021
Max Planck Institute for Metabolism Research
2021
NIHR Birmingham Biomedical Research Centre
2021
University Hospitals Birmingham NHS Foundation Trust
2021
Mayo Clinic in Arizona
2021
Dialyse Centrum Groningen
2020
Diakonie Deutschland
2019
Universitat Politècnica de Catalunya
2019
A summary is presented of the statistical mechanical theory learning a rule with neural network, rapidly advancing area which closely related to other inverse problems frequently encountered by physicists. By emphasizing relationship between networks and strongly interacting physical systems, such as spin glasses, authors show how has provided workshop in develop new, exact analytical techniques.
Adrenal tumors have a prevalence of around 2% in the general population. Adrenocortical carcinoma (ACC) is rare but accounts for 2-11% incidentally discovered adrenal masses. Differentiating ACC from adrenocortical adenoma (ACA) represents diagnostic challenge patients with incidentalomas, tumor size, imaging, and even histology all providing unsatisfactory predictive values.Here we developed novel steroid metabolomic approach, mass spectrometry-based profiling followed by machine learning...
We propose a new matrix learning scheme to extend relevance vector quantization (RLVQ), an efficient prototype-based classification algorithm, toward general adaptive metric. By introducing full of factors in the distance measure, correlations between different features and their importance for can be taken into account automated, metric adaptation takes place during training. In comparison weighted Euclidean used RLVQ its variations, is more powerful represent internal structure data...
We describe and compare several post-correlation radio frequency interference (RFI) classification methods. As data sizes of observations grow with new improved telescopes, the need for completely automated, robust methods RFI mitigation is pressing. investigated find that, sets we used, most accurate among them SumThreshold method. This a method formed from combination existing techniques, including way thresholding. iterative estimates astronomical signal by carrying out surface fit in...
Adrenal aldosterone excess is the most common cause of secondary hypertension and associated with increased cardiovascular morbidity. However, adverse metabolic risk in primary aldosteronism extends beyond hypertension, rates insulin resistance, type 2 diabetes, osteoporosis, which cannot be easily explained by excess.We performed mass spectrometry-based analysis a 24-hour urine steroid metabolome 174 newly diagnosed patients (103 unilateral adenomas, 71 bilateral adrenal hyperplasias)...
Cross-sectional imaging regularly results in incidental discovery of adrenal tumours, requiring exclusion adrenocortical carcinoma (ACC). However, differentiation is hampered by poor specificity characteristics. We aimed to validate a urine steroid metabolomics approach, using profiling as the diagnostic basis for ACC.
A new learning algorithm for neural networks of spin glass type is proposed. It found to relax exponentially towards the perceptron optimal stability using concept adaptive learning. The patterns can be presented either sequentially or in parallel. prove convergence given and method's performance studied numerically.
Background and objectives For our understanding of the pathogenesis rheumatoid arthritis (RA), it is important to elucidate mechanisms underlying early stages synovitis. Here, synovial cytokine production was investigated in patients with very arthritis. Methods Synovial biopsies were obtained from at least one clinically swollen joint within 12 weeks symptom onset. At an 18-month follow-up visit, who went on develop RA, or whose spontaneously resolved, identified. Biopsies also RA longer...
We study on-line gradient-descent learning in multilayer networks analytically and numerically. The training is based on randomly drawn inputs their corresponding outputs as defined by a target rule. In the thermodynamic limit we derive deterministic differential equations for order parameters of problem which allow an exact calculation evolution generalization error. First consider single-layer perceptron with sigmoidal activation function rule network same architecture. For this model...
In recent years, a wealth of dimension-reduction techniques for data visualization and preprocessing has been established. Nonparametric methods require additional effort out-of-sample extensions, because they provide only mapping given finite set points. this letter, we propose general view on nonparametric dimension reduction based the concept cost functions properties data. Based principle, transfer to explicit mappings manifold such that direct extensions become possible. Furthermore,...
Discriminative vector quantization schemes such as learning (LVQ) and extensions thereof offer efficient intuitive classifiers based on the representation of classes by prototypes. The original methods, however, rely Euclidean distance corresponding to assumption that data can be represented isotropic clusters. For this reason, methods more general metric structures have been proposed, relevance adaptation in generalized LVQ (GLVQ) matrix GLVQ. In these approaches, parameters are learned...
Abstract Context Urine steroid metabolomics, combining mass spectrometry-based profiling and machine learning, has been described as a novel diagnostic tool for detection of adrenocortical carcinoma (ACC). Objective, Design, Setting This proof-of-concept study evaluated the performance urine metabolomics postoperative recurrence after microscopically complete (R0) resection ACC. Patients Methods 135 patients from 14 clinical centers provided samples, which were analyzed by gas...
We study layered neural networks of rectified linear units (ReLU) in a modelling framework for stochastic training processes. The comparison with sigmoidal activation functions is the center interest. compute typical learning curves shallow K hidden matching student teacher scenarios. systems exhibit sudden changes generalization performance via process unit specialization at critical sizes set. Surprisingly, our results show that behavior ReLU qualitatively different from activations. In...
Abstract Parkinson’s disease (PD) is characterized by a progressive loss of dopaminergic neurons in the substantia nigra. Recent literature has proposed two subgroups PD. The “body-first subtype” associated with prodrome isolated REM-sleep Behavior Disorder (iRBD) and relatively symmetric brain degeneration. “brain-first suggested to have more asymmetric degeneration prodromal stage without RBD. This study aims investigate difference symmetry pattern presumed body brain-first PD subtypes. We...