Institutional Repository of Vilnius University
Atomic chess ai development using alpha-beta pruning and machine learning /
Abstract
dc:descriptionThis paper aims to extend the knowledge of artificial intelligence (AI) systems for atomic chess, a chess variant known for its unique rule of explosive piece captures. An AI system is developed and evaluated, integrating alpha-beta search algorithm, related optimizations, and neural networks for positional analysis. The creation of an atomic chess engine is detailed, along with the application of minimax tree search acceleration techniques and the design of a handcrafted position’s evaluation function. The integration of efficiently updatable neural networks (NNUE) with the engine is explored, alongside the impact that various training dataset parameters have on the engine’s playing strength. The findings reveal a notable performance improvement when using NNUE compared to handcrafted evaluation (HCE). Key conclusions affirm the transferability of standard chess AI techniques to atomic chess, the importance of piece mobility and values in static evaluation, and the effectiveness of smaller NNUE architectures.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Cibulskis, Martynas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192846153&prefLang=en_US