Globoko učenje in igra dama
diplomsko delo
Abstract
V zaključnem delu smo zasnovali računalniški program AlphaLady, ki se je sposoben naučiti igranja igre dama brez vnosa človeškega znanja. Za dosego tega smo uporabili vzpodbujevalno učenje, drevesno preiskovanje Monte Carlo in globoke konvolucijske mreže za ocenitev posameznih stanj v igri. Predstavili smo programe Alpha Go, AlphaGo Zero in AlphaZero, na podlagi katerih je zasnovan naš program. Opisali smo uporabljeno ogrodje in teoretično ozadje uporabljenih pristopov. Uspelo nam je naučiti 9 različic programa, pri čemer je vsaka naslednja različica enakovredna ali boljša kot prejšnja.
Keywords
umetna inteligenca;globoko učenje;konvolucijska nevronska mreža;drevesno preiskovanje Monte Carlo;vzpodbujevalno učenje;igra dama;diplomske naloge;
Data
| Language: | Slovenian |
|---|---|
| Year of publishing: | 2019 |
| Typology: | 2.11 - Undergraduate Thesis |
| Organization: | UM FERI - Faculty of Electrical Engineering and Computer Science |
| Publisher: | [J. Popič] |
| UDC: | 004.8(043.2) |
| COBISS: |
22848534
|
| Views: | 907 |
| Downloads: | 205 |
| Average score: | 0 (0 votes) |
| Metadata: |
|
Other data
| Secondary language: | English |
|---|---|
| Secondary title: | Deep Learning and the game of Checkers |
| Secondary abstract: | In the thesis, we designed a computer program AlphaLady, which is capable of learning to play the game of checkers without human knowledge. To achieve this we have used reinforcement learning, Monte Carlo tree search and deep convolutional neural network for evaluating board positions. Our program is based on the introduced programs Alpha Go, AlphaGo Zero and AlphaZero. We described a framework that was used for implementation and theoretical background of used approaches. We managed to train 9 versions of our program, with each successive version being equal or better than the previous one. |
| Secondary keywords: | artificial intelligence;deep learning;convolutional neural network;Monte Carlo tree search;reinforcement learning;checkers; |
| Type (COBISS): | Bachelor thesis/paper |
| Thesis comment: | Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije |
| Pages: | XVII, 50 str. |
| ID: | 11205917 |
Recommended works:
Globoko učenje in igra dama
2019,
diplomsko delo
Učenje igranja igre štiri v vrsto z uporabo globokega spodbujevalnega učenja
2021,
magistrsko delo
FighterZero
2018,
pristop samo-igranja za učenje igranja pretepaške igre z globokim spodbujevalnim učenjem
Napovedovanje geoprostorskih rastrskih podatkov s konvolucijskimi nevronskimi mrežami
2020,
diplomsko delo
Monte Carlo tree search strategies
2018,
no subtitle data available