Topic
Recurrent neural networks
Recurrent networks run through three separate research lines here. For mine workers, a recurrent network estimated hearing threshold shift from noise exposure, with a comparison of optimisation methods for training it. For seasonal climate forecasting, an RNN was tested against a CNN at predicting 2m temperature up to 52 weeks ahead. And in the MSc work, an echo state network, a recurrent model whose internal weights are never trained, was compared against trained convolutional RNN, LSTM and GRU models at learning level set image segmentation.
3 research · 3 publications · Wikidata Q1457734
Research
- Echo state networks for level set image segmentationMSc research at Wits: can recurrent models learn the iterative evolution of a level set segmentation? Echo state networks against trained RNN, LSTM and GRU.
- Machine learning for seasonal climate forecastingIBM Research work on forecasting 2m temperature and precipitation weeks to months ahead with machine learning, published at NeurIPS 2020 CCAI and EGU 2022.
- Machine learning for hearing loss risk in mine workersTwo 2019 IFAC papers: a recurrent network estimating hearing threshold shift in mine workers, and a classifier based noise policy advising system.
Publications
- Learning Level Set Method by Echo State Network for Image SegmentationStudies echo state networks as a cheaper alternative to recurrent networks trained by backpropagation, applied to learning variational level set image segmentation as a spatiotemporal, data driven method. Five convolutional architectures were compared (ESN, RNN, GRU, LSTM and a 3D CNN) on four datasets; the GRU and LSTM variants performed best. The ESN performed poorly, which the dissertation attributes largely to the reservoir's leaking rate and spectral radius.
- Long-Range Seasonal Forecasting of 2m-Temperature with Machine LearningTests whether two machine learning models, a convolutional and a recurrent neural network, can beat climatology when forecasting 2 m temperature up to 52 weeks ahead at six locations. The models improved on climatology up to 30 weeks lead time by correlation and up to 52 weeks by RMSE skill score, but only at some locations, and the authors note further work is needed for the models to add value in the tropics.
- Mine workers threshold shift estimation via optimization algorithms for deep recurrent neural networksApplies recurrent neural networks to estimate hearing threshold shift in mine workers, and compares optimisation methods for training them. The adaptive subgradient method (Adagrad) was preferred for its fast convergence, and the network predicted threshold shift with 95% accuracy. The authors suggest the results could support an early intervention and monitoring system for mines.
Related topics
Subjects that share work with this one. Those with their own page are linked.
- Machine learning
- Deep learning
- Climate risk
- Computer vision
- Echo state networks
- Image segmentation
- Level set method
- Noise induced hearing loss
- Occupational health and safety
- Reservoir computing
- Seasonal forecasting
- Geospatial machine learning