Randomizirana metoda optimizacije ADAM

master's thesis

• Title (english): Randomized optimization method ADAM
• Author: Igor Marić
• Mentor: Luka Grubišić (mentor)
• Committee member: Luka Grubišić (predsjednik povjerenstva)
• Committee member: Zvonimir Bujanović (član povjerenstva)
• Committee member: Mladen Vuković (član povjerenstva)
• Committee member: Igor Ciganović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2023-05-04, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

U ovom radu obrađene su stohastičke gradijentne metode. Analizu smo većinom sproveli na metodi linearne regresije. Svi opimizatori su zbog te svrhe implementirani. Iz primjera smo vidjeli da se u preciznosti i efikasnosti najviše iskazala metoda adaptivnog reduciranja varijance Adam. Potom, opisali smo biblioteke automatskog deriviranja koda Jax i Autograd te na osnovu njih uvidjeli znatnu olakšicu u primjenama optimizatora. Naposljetku, implementirali smo algoritam za restauraciju slika naziva HRMC.

Abstract (english)

This thesis deals with stochastic gradient methods. We mostly conducted the analysis using the linear regression method. All optimizers are implemented for this purpose. From the examples, we saw that the method of adaptive variance reduction Adam was the most accurate and efficient. Then, we described the automatic code derivation libraries Jax and Autograd, and based on them, we saw a significant improvement in optimizer applications. Finally, we implemented an image restoration algorithm called HRMC.

• Keywords: stohastičke gradijentne metode; metoda linearne regresije; metoda adaptivnog reduciranja varijance Adam; algoritam za restauraciju slika HRMC; Jax; Autograd
• Keywords (english): stochastic gradient methods; linear regression method; method of adaptive variance reduction Adam; Jax; Autograd; mage restoration algorithm HRMC
• Language: croatian
• URN:NBN: urn:nbn:hr:217:188192
• Study programme: Title: Computer Science and Mathematics; Study programme type: university; Study level: graduate; Academic / professional title: sveučilišni magistar računarstva i matematike (sveučilišni magistar računarstva i matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-02-09 12:40:52