Lecture 14.1: deflected gradient methods I - Conjugate Gradient
Aggregazione dei criteri
Towards the very-large-scale: limited-memory quasi-Newton (sketch). A different (cheaper) direction modification: deflection. Deflection-type methods I: conjugate gradient, the "father" of all deflection-tipe methods. Different formulae, theoretical results (sketch). Different \beta formulae, theoretical results (sketch). MATLAB implementation of the conjugate gradient. Behaviour on relevant test cases, impact of algorithmic parameters (\beta formula, restarts). Take away: interesting in theory, but practice is hit-and-miss.