Deep learning for general game playing with Ludii and Polygames

Abstract

Combinations of Monte-Carlo tree search and Deep Neural Networks, trained through self-play, have produced state-of-the-art results for automated game-playing in many board games. The training and search algorithms are not game-specific, but every individual game that these approaches are applied to still requires domain knowledge for the implementation of the game’s rules, and constructing the neural network’s architecture – in particular the shapes of its input and output tensors. Ludii is a general game system that already contains over 1,000 different games, which can rapidly grow thanks to its powerful and user-friendly game description language. Polygames is a framework with training and search algorithms, which has already produced superhuman players for several board games. This paper describes the implementation of a bridge between Ludii and Polygames, which enables Polygames to train and evaluate models for games that are implemented and run through Ludii. We do not require any game-specific domain knowledge anymore, and instead leverage our domain knowledge of the Ludii system and its abstract state and move representations to write functions that can automatically determine the appropriate shapes for input and output tensors for any game implemented in Ludii. We describe experimental results for short training runs in a wide variety of different board games, and discuss several open problems and avenues for future research.

A Convolutional Neural Network Approach to General Game Playing

“…Moreover, we also conduct a direct comparison of our agent with GAZ. We do not compare with agents built on other systems, such as Ludii [10]. On the one hand, cross-platform competitions are not supported by the current system.…”

“…The neural networks in the above three works only employed fully connected layers. Most recently, based on another GGP system, Ludii [22], and DRL framework, polygames [23], CNNs were employed in GGP [10]. It achieves very good performance on 15 board games, but is still limited to 2D, two-player, asynchronous games.…”

“…It removes the limitation on game type and can automatically design the neural networks according to the number of game propositions [8]. It is the first deep reinforcement learning algorithm that outperforms the UCT benchmark in all games [7,9,10], except in multiplayer cooperative games like Babel. However, unlike Al-phaZero, GAZ employs fully connected networks instead of convolutional neural networks (CNNs).…”

Fast and Knowledge-Free Deep Learning for General Game Playing (Student Abstract)

“…A clone of Al-phaZero, based on GDL and aimed to resolve some of these restrictions, was presented in (Goldwaser and Thielscher 2020). MCTS-based DRL approach for Ludii GGP system, created via a bridge to Polygames can be found in (Soemers et al 2021). Deep Reinforcement Learning using only value networks combined with a variant of Unbounded Minimax is described in (Cohen-Solal and Cazenave 2023).…”

Developing and Assessing a Human-Understandable Metric for Evaluating Local Interpretable Model-Agnostic Explanations

“…Artificial intelligence (AI) systems have pervaded numerous fields, ranging from facial recognition [1] and gaming [2,3], to text analysis [4][5][6] and natural language processing in consumer devices [7]. They have also become invaluable tools in scientific research for prediction, simulation, and exploration [8][9][10].…”