Demonstracijski okvir za diferencijabilno programiranje

undergraduate thesis

• Title (english): Demonstration framework for differentiable programming
• Author: Jakov Novak
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Ivan Sabolić (član povjerenstva)
• Committee member: Iva Sović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-01, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Kako neuralne mreže postaju sve složenije i popularnije, ključno je razumjeti mehanizme koji omogućuju njihovo efikasno učenje. U ovom radu istražujemo matematičke osnove automatskog diferenciranja, što je temeljna tehnika za treniranje neuralnih mreža. Detaljno razmatramo teorijske koncepte koji stoje iza automatskog diferenciranja, uključujući derivacije i gradijente, te njihovu primjenu u optimizaciji modela strojnog učenja. Osim teorijskog pregleda, rad obuhvaća i praktičnu implementaciju jednostavnog sustava za automatsko diferenciranje. Pomoću ovog sustava demonstriramo njegovu primjenu kroz nekoliko programskih primjera. Na kraju, razvili smo klasifikator slika koristeći vlastiti sustav za automatsko diferenciranje, čime smo pokazali praktičnu korisnost i efikasnost razvijenog rješenja u stvarnim zadacima strojnog učenja.

Abstract (english)

As neural networks become increasingly complex and popular, understanding the mechanisms that enable their efficient learning is crucial. This paper explores the mathematical foundations of automatic differentiation, a fundamental technique for training neural networks. We thoroughly examine the theoretical concepts behind automatic differentiation, including derivatives and gradients, and their application in optimizing machine learning models. In addition to the theoretical overview, the paper includes a practical implementation of a simple automatic differentiation system. We demonstrate the system's application through several programming examples. Finally, we developed an image classifier using our custom automatic differentiation system, showcasing the practical utility and efficiency of the developed solution in real-world machine learning tasks.

• Keywords: autodiff; c++; automatsko diferenciranje; strojno učenje; duboko učenje
• Keywords (english): autodiff; c++; automatic differentiation; machine learning; deep learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:739472
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 21:41:21