UMass Machine Learning and Friends Lunch | Main / Scalable And Robust Sparse Subspace Clustering

Abstract: Sparse subspace clustering (SSC) is a popular method in machine learning and computer vision for clustering high-dimensional data points that lie near a union of low-dimensional linear or affine subspaces. Using a two-step approach, the first step involves representing each data point as a linear combination of all other data points, so-called self-expressiveness property, to form an undirected similarity graph. Spectral clustering is then applied to produce the final segmentation and infer the underlying subspaces. The sparse optimization program in the first step of SSC is typically solved by the alternating direction method of multipliers (ADMM) that scales cubically with the number of data points. In addition, the process of optimal parameter selection for ADMM requires a significantly increased amount of computational time. Orthogonal matching pursuit (OMP) has been used as a more efficient alternative; however, OMP is incapable of handling affine subspaces and the choice of sparsity parameter notably impacts the accuracy of SSC.

Bio: Farhad Pourkamali Anaraki is an Assistant Professor of Computer Science at UMass Lowell. He received his PhD from University of Colorado Boulder under the supervision of Prof. Stephen Becker in 2017. His current research focuses on theoretical foundations of modern data science, and developing efficient and robust machine learning algorithms. He also works on extending the use of machine learning algorithms to other domains, including åreliability analysis of critical infrastructures in Civil Engineering and advanced manufacturing in Mechanical Engineering.