Razvoj arhitekture umjetne neuronske mreže evolucijom

undergraduate thesis

• Title (english): Evolution of Artificial Neural Network Architecture
• Author: Matija Bačić
• Mentor: Marko Čupić (mentor)
• Committee member: Marko Čupić (predsjednik povjerenstva)
• Committee member: Rudolf Lovrenčić (član povjerenstva)
• Committee member: Karlo Knežević (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Konektivistička paradigma u umjetnoj inteligenciji danas je najrašireniji pristup. Pojam neuronske mreže uveo je revoluciju u računarsku znanost jer sada možemo riješiti puno kompleksnije probleme na jednostavniji način - treniranjem. Ideja više nije unijeti sve poznato znanje u računalo, već pustiti neuronsku mrežu da sama zaključuje kako se ponašati u još neviđenim situacijama. Dva su načina za to postići: matematički orijentirani način propagacije unazad te neuroevolucija. Zbog svoje jednostavnosti u matematičkom pogledu, nauroevolucija ima značajnu popularnost u svijetu umjetne inteligencije. Međutim, veliki problem predstavlja odabir topologije neuronske mreže koju paniramo trenirati. Iako postoje mnoge dobro poznate topologije za modeliranje određenih problema, postavlja se pitanje bismo li dobili napredak evoluiranjem cijele topologije, a ne samo težine spojeva mreže. Algoritam neuroevolucije s povećavanjem topologija (NEAT) upravo daje odgovor na to pitanje.

Abstract (english)

The connectivism in artificial intelligence is the most widespread paradigm today. The concept of neural network has revolutionized computer science since it enables solving complex problems in a simpler way – by training. The idea is not to enter all the available knowledge into the computer, rather let the artificial neural network decide how to behave in unprecedented situations. There are two ways to achieve that: a mathematically oriented way of backwards propagation and neuroevolution. Because of its mathematical simplicity, nauroevolution has gained significant popularity in the world of artificial intelligence. However, a big problem is choosing the neural network topology beforehand. Although there are many good already known topologies for modeling specific problems, question arises would we gain significant improvement by evolving network topologies along with connection weights. The neuroevolution with augmenting topologies (NEAT) provides the answer to the posed question.

• Keywords: neuroevolucija; topologija; neuronska mreža; NEAT; evolucijski algoritmi
• Keywords (english): neuroevolution; topology; neural network; NEAT; evolutionary algorithms
• Language: english
• URN:NBN: urn:nbn:hr:168:988102
• 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: 2023-01-14 08:21:22