Proper Orthogonal Decomposition (POD) is a dimensionality reduction technique that is used to capture the energetically dominant features of datasets, known as eigenfeatures or POD modes. The datasets involved are typically large, but correlated and have a relatively small number of dominant POD modes. Randomised algorithms have been used to efficiently compute POD of large datasets using Singular Value Decomposition (SVD). They can be broadly classified as: random projection and random sampling techniques. Random sampling based algorithms are advantageous as they retain or increase the sparsity of the matrix and are easily amenable for incremental improvement in accuracy. Some of the existing column/row sampling based algorithms either are not computationally efficient or approximate the modes with a large error. In this seminar, we propose an iterative sampling algorithm that achieves excellent accuracy and gives significant improvement in runtime compared to SVD. A posteriori estimation of error in the subspaces spanned by the modes is computed using Wedin’s theorem. This error measure acts as a termination condition for the approximation. For large datasets that do not fit in the RAM, we propose an incremental algorithm that requires only one pass over the data. We use this algorithm in conjunction with the iterative sampling algorithm to compute the POD modes of large datasets.