Data & Algebra Seminar

The canonical polyadic decomposition (CPD) of a tensor into a minimal sum of rank-one terms plays a crucial role in many signal processing applications, enabling unique recovery of underlying components. However, computing the CPD of a measured signal tensor requires first finding a best low-rank tensor approximation, a task that is both ill-posed and NP-hard in general. This talk will show that, despite these challenges, low-rank tensor approximation is well-posed for many tensors arising in applications. Additionally, we discuss algebraic algorithms for computing the CPD in such cases.  This talk will occur through zoom (https://utexas.zoom.us/j/94485676974).  It will also be broadcast live in PMA 9.166, where an in-person viewing party will assemble.