Odkrivanje znanj igre Blackjack z metodami spodbujevanega učenja

FRANKO GORIŠEK, ŽIGA (author), Sadikov, Aleksander (mentor)

Skozi študijski proces, sem pridobil veliko znanja ter zanimanja na področjih, kot so algoritmi, optimizacija, stave, finance ter umetna inteligenca. Tako sem razvil željo, da bi se specializiral na področju Globokega učenja (Deep learning). Želel sem doseči stopnjo znanja in razumevanja kako naučeno specializacijo izkoristiti z najnovejšimi tehnologijami strojnega učenja. Izredno me je tudi prevzela zgodba o MIT študentih, ki so odkrili in izkoristili luknjo v igri Black jack, s katere so si odprli dobičkonosno igro. Zato sem se odločil narediti sistem, ki se nauči odkriti zakonitosti igre Black jack, se tako naučiti tehnike DQN in poskušati izboljšati osnovo njihovega sistema.

• Language: Slovenian
• Keywords: DQN, Nevronske mreže, Black jack, delivec, igralec, model
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2018
• PID: 20.500.12556/RUL-102501
• Publication date in RUL: 31.08.2018

Secondary language

• Language: English
• Title: Learning Blackjack using reinforcement learning

Abstract

Through the study process, I gained a lot of knowledge and interests in fields such as algorithms, optimization, betting, finance and artificial intelligence. So I developed a desire to specialize in Deep Learning. I wanted to achieve a level of knowledge and understanding of how to use the learned specialization with the latest machine learning technologies. I was also extremely impressed by the story about MIT students who discovered and used the Black jack hole from which they opened a profitable game. So I decided to make a system that learns how to discover the legitimacy of the Black jack game, to learn DQN techniques and try to improve the basis of their system with it.

• Keywords: DQN, Neural network, Black jack, dealer, player, model