Anh Huy Phan
Skolkovo Institute of Science and TechnologyHome page
Conduct research on the following main projects
- Tensor decompositions (TDs) and novel applications: Deal with most challenging problems in tensor decompositions, e.g., high computational cost of algorithms, the second-order optimization method, a measure of performance of algorithms, massive tensor decomposition, tensor deflation.
- Develop robust algorithms with low computational costs for various low-rank tensor decompositions of large-scale data, including the Candecomp/Parafac, Tucker tensor decompositions, low-rank tensor deconvolution, tensor diagonalization, Kronecker tensor decomposition, tensor deflation, feature extraction for multiway data.
- We contributed state-of-the-art algorithms to the tensor analysis, especially for the Candecomp/Parafac tensor decomposition.
- Blind Sources Separation. We developed the algorithms for blind source separation based on independent component analysis, nonnegative matrix factorization and tensor network decomposition. Our methods consist of two steps: tensorization of the mixture signals to yield tensors, low-rank tensor decomposition to retrieve the hidden sources.
- Brain Computer interface (BCI): we applied tensor decomposition to extract relevant features for EEG signals in BCI system.
- Nonlinear system identification based on multivariate polynomial regression, Volterra-type models, in which parameters are represented in Tensor network format. Our models can be applied to regression, discriminant analysis, support vector machines, deep learning, data fusion and feature combination.
- Early detection of Alzheimer disease
- Deconvolution of large-scale Calcium imaging data
- Coding of faces by components with complexity constraints
Publications since 2016