A5013354515- Time Series Analysis and Forecasting
- Insurance, Mortality, Demography, Risk Management
- Machine Learning in Healthcare
- Markov Chains and Monte Carlo Methods
- Statistical Methods and Inference
- Astrophysics and Cosmic Phenomena
- Radio Astronomy Observations and Technology
- Advanced Statistical Methods and Models
- Spectroscopy and Chemometric Analyses
- Geophysical Methods and Applications
- Pulsars and Gravitational Waves Research
- Neural Networks and Applications
- Fault Detection and Control Systems
- Gamma-ray bursts and supernovae
- Image Enhancement Techniques
- Skin Protection and Aging
- Imbalanced Data Classification Techniques
- Galaxies: Formation, Evolution, Phenomena
- Textile materials and evaluations
- Stellar, planetary, and galactic studies
- Cosmology and Gravitation Theories
- 3D Shape Modeling and Analysis
- Astronomy and Astrophysical Research
- Advanced Vision and Imaging
- Oil and Gas Production Techniques
Microsoft Research (United Kingdom)
2024
University of Manchester
2019-2023
South African Radio Astronomy Observatory
2017-2019
African Institute for Mathematical Sciences
2017-2019
North-West University
2017
We compute the Bayesian evidence for models considered in main analysis of Planck cosmic microwave background data. By utilizing carefully defined nearest-neighbor distances parameter space, we reuse Monte Carlo Markov chains already produced inference to Bayes factors $B$ many different model-data set combinations. The standard 6-parameter flat cold dark matter model with a cosmological constant ($\mathrm{\ensuremath{\Lambda}}\mathrm{CDM}$) is favored over all other considered, curvature...
In this paper, we present a method for computing the marginal likelihood, also known as model likelihood or Bayesian evidence, from Markov Chain Monte Carlo (MCMC), other sampled posterior distributions. order to do this, one needs be able estimate density of points in parameter space, and can challenging high numbers dimensions. Here analysis, where obtain using $k$th nearest-neighbour distances Mahalanobis distance metric, under assumption that chain (thinned if required) are independent....
Point source detection at low signal-to-noise is challenging for astronomical surveys, particularly in radio interferometry images where the noise correlated. Machine learning a promising solution, allowing development of algorithms tailored to specific telescope arrays and science cases. We present DeepSource - deep solution that uses convolutional neural networks achieve these goals. enhances Signal-to-Noise Ratio (SNR) original map then dynamic blob detect sources. Trained tested on two...
Upcoming synoptic surveys are set to generate an unprecedented amount of data. This requires automatic framework that can quickly and efficiently provide classification labels for several new object challenges. Using data describing 11 types variable stars from the Catalina Real-Time Transient Surveys (CRTS), we illustrate how capture most important information computed features describe detailed methods robustly use Information Theory feature selection evaluation. We apply three Machine...
ABSTRACT The accurate automated classification of variable stars into their respective subtypes is difficult. Machine learning–based solutions often fall foul the imbalanced learning problem, which causes poor generalization performance in practice, especially on rare star subtypes. In previous work, we attempted to overcome such deficiencies via development a hierarchical machine classifier. This ‘algorithm-level’ approach tackling imbalance yielded promising results Catalina Real-Time...
We tackle the problem of highly-accurate, holistic performance capture for face, body and hands simultaneously. Motion-capture technologies used in film game production typically focus only on or hand independently, involve complex expensive hardware a high degree manual intervention from skilled operators. While machine-learning-based approaches exist to overcome these problems, they usually support single camera, often operate part body, do not produce precise world-space results, rarely...
We present a method for prediction of person's hairstyle from single image. Despite growing use cases in user digitization and enrollment virtual experiences, available methods are limited, particularly the range hairstyles they can capture. Human hair is extremely diverse lacks any universally accepted description or categorization, making this challenging task. Most current rely on parametric models at strand level. These approaches, while very promising, not yet able to represent short,...