Geometry and Computer Vision with Machine Learning for Scene Analysis (linked to ARIES 32367)

The project will develop methods for using Machine Learning tools in Computer Vision, particularly emphasizing methods involving the use of geometrical data analysis. The project will also focus on the use of Markov Random Field (MRF) techniques and Convolutional Neural Networks (CNNs) as appropriate machine learning tools. One of the themes is how MRFs and CNNs can be amalgamated to get the best results from both techniques. MRFs and CNNs or Recursive Neural Networks (RNNs) can be amalgamated into one structure, and trained together to give better results on problems such as image segmentation. This theme has been investigated by my Data61 student Ajanthan Thalaisingham in collaboration with Oxford University, and has been shown to be promising. This theme will be developed along with the study of data that naturally sits in well-defined geometric formations, such as Riemannian manifolds, which has been shown to be very advantageous for refining the techniques.