27 Moucheng Xu Computer Vision News His second work of the thesis defines a new formulation of pseudo labelling as the expectation maximization algorithm, providing an interpretable perspective of the empirical effectiveness of the pseudo labelling in semisupervised learning (Fig. 2). Based on this newly proposed formulation, his work extends the original pseudo labelling towards its generalization and presents an approximation of its generalization using a variational inference. This work resulted in a publication at MICCAI 2022, which was fortunately shortlisted for the Young Scientist Award, and another patent application. The final work of his thesis leverages the latest advancements in variational unsupervised clustering techniques with discrete priors and demonstrates its first application on model parameter estimation of MRI signals. This pioneering approach challenges the traditional assumption that all of the voxels are treated independently in MRI parameter estimations. Additionally, the new approach enables a new generation of dMRI and qMRI with enhanced anatomical structures while reducing noises. For more details, reach out to him.
RkJQdWJsaXNoZXIy NTc3NzU=