Résumé | Linear algebra is behind most scientific analytic computations, including numerical methods, statistics,
physics and data mining. On the other hand, database systems provide extensive features to manage and
query data, but not to analyze it. The integration of linear algebra algorithms into a database system is
difficult due to the relational model discrete set foundation, which in consequence hinders efficient array
processing. In this talk, we discuss our progress towards solving two prominent linear algebra problems
on large data sets: singular value decomposition and least squares regression. We present a common
matrix foundation to summarize large data sets with an outer product and exploiting such summarization
we discuss how to efficiently solve linear algebra problems with iterative methods. |