Metode učenja neuronske mreže za igru zmija

undergraduate thesis

• Title (english): Methods for learning neural networks for the Snake game
• Author: Nikola Kušen
• Mentor: Marko Đurasević (mentor)
• Committee member: Marko Đurasević (predsjednik povjerenstva)
• Committee member: Josip Hrvatić (član povjerenstva)
• Committee member: Ana Barić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-06-27, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U radu je opisan računalni model neuronske mreže, njegove primjene te načini učenja
modela. Analizirano je podržano učenje koristeći genetski algoritam te nadzirano učenje
gradijentnim spustom i algoritmom propagacije pogreške unazad (eng. backpropaga-
tion). Tehnike su implementirane i isprobane kroz učenje mreže kako bi igrala video igru
zmija. Rezultati pokazuju da je genetski algoritam pronašao bolju strategiju za igru od
mreža koje su napravljene nadziranim učenjem uspoređujući svoje poteze s algoritmom
pretraživanja u širinu (BFS) ili jednostavnim algoritmom razvijenim u sklopu rada. Una-
toč tome, nadzirano učenje pokazuje potencijal u smislu prilagodbe algoritmu od kojeg
uči, pa bi dobrim odabirom tog algoritma mreža mogla imati još bolje rezultate.

Abstract (english)

This paper describes a computational model of neural networks, its applications, and
ways of training the model. It also analyses reinforcement learning through a genetic
algorithm and supervised learning using gradient descent and the backpropagation al-
gorithm. The techniques are implemented and tested by training a neural network to
play the video game Snake. The results show that the genetic algorithm found a better
strategy than networks trained by comparing their moves to breadth-first search (BFS)
or a simple algorithm developed as a part of this study. In spite of that, supervised learn-
ing shows potential by adapting well to an algorithm it learns from, meaning that a good
choice of that algorithm could yield even better results.

• Keywords: neuronske mreže; podržano učenje; genetski algoritam; nadzirano učenje; gradijentni spust; propagacija pogreške unazad
• Keywords (english): neural networks; reinforcement learning; genetic algorithm; supervised learning; gradient descent; backpropagation
• Language: croatian
• URN:NBN: urn:nbn:hr:168:915815
• 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 21:16:03