Numerična optimizacija z metodami usmerjenega spusta : delo diplomskega seminarja

Ribič, Brina (author), Grošelj, Jan (mentor)

V delu se ukvarjamo s problemom iskanja najmanjše vrednosti funkcije z orodji numerične optimizacije. Obravnavamo iterativne algoritme oziroma metode, za katere si želimo, da z visokim redom konvergence zanesljivo privedejo do rešitve. Na začetku izpeljemo metodo najstrmejšega spusta, ki pa ima v splošnem kvečjemu linearen red konvergence in se v praksi redko uporablja. Nato analiziramo Newtonovo metodo z višjim redom konvergence, a je njena pomanjkljivost v tem, da je za izvedbo treba računati Hessejevo matriko funkcije. Predstavimo še BFGS metodo, ki ima superlinearen red konvergence, hkrati pa ni računsko zahtevna. V zaključnem poglavju primerjamo BFGS metodo in metodo najstrmejšega spusta na primeru CAPM modela, pri katerem z linearno regresijo na podlagi podatkov preteklih let iščemo beta koeficient izbranega podjetja. Ugotovimo, da je BFGS metoda bolj zanesljiva in učinkovita kot metoda najstrmejšega spusta.

• Language: Slovenian
• Keywords: neomejena optimizacija, metode usmerjenega spusta
• Work type: Bachelor thesis/paper
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2022
• PID: 20.500.12556/RUL-139563
• UDC: 519.6
• COBISS.SI-ID: 120694787
• Publication date in RUL: 04.09.2022

Secondary language

• Language: English
• Title: Numerical optimization with line search methods

Abstract

In this work we study the problem of finding the minimum value of a function using numerical optimization tools. We seek for algorithms which have high rate of convergence and efficiently find solution. We start by deriving the steepest descent method which in general has at most linear rate of convergence and is rarely used in practice. Later on we analyze Newton's method with a higher rate of convergence, but its main drawback is that it requires the computation of the Hessian matrix of the function. We also introduce the BFGS method which has superlinear rate of convergence and has low computitonal complexity. At the end of this work we compare BFGS method and the steepest descent method on CAPM model. We use linear regression, where we search for beta coeficient based on data from previous years. We conclude that the BFGS method is more reliable and efficient than the steepest descent method.

• Keywords: unconstrainted optimization, line search methods