Abstract:
|
T.W. Anderson’s 1951 Annals of Statistics paper based on his 1948 Princeton thesis on multivariate regression with linear constraints on the coefficient matrix has played an influential role in Statistics, Econometrics and Machine learning. An important aspect of this thesis is that these constraints are not known a priori and are solely determined by the data at hand. A complementary aspect of this idea is that the coefficient matrix is of reduced-rank, a term coined by Izenman(1975). Because of this duality, the method has a wider appeal. It has historical connections to topics such as latent variables, discriminant analysis etc. With the focus on big data now, the method along with sparseness provides efficient way to handle large scale modeling and its application in modeling chronological data has been widely recognized. In this talk, I will provide a summary of major developments in the past seventy years.
|