AbstractIn the current state of the video game productions, most of the video game levels are created by the human operators working as level designers. This manual process is not only time-consuming and resource-intensive but also hard to guarantee uniform quality in the contents created by the level designers. One way to address this issue is to use computer-assisted level design techniques. In this paper, we have proposed a novel framework for computer-assisted video game level design that leverages neural networks, particularly generative adversarial networks (GANs) and autoencoders. The general idea is to learn over a dataset of high-quality levels and subsequently improve the ones created by the level designers. The proposed method is independent of the graphical dimensionality of the game and will work for 2D and 3D games in general. The autoencoder is used to create an intermediate representation of the level that is itself changed using the backpropagation technique according to the feedback obtained by feeding the output of the autoencoder to the discriminator component of the GAN. After performing a series of evaluations on the proposed framework and by automatically improving a series of purposefully corrupted game levels, the results demonstrate a noticeable improvement compared with the usage of simple autoencoders used to improve the video game levels in the previous researches.
Growing concerns regarding the operational usage of AI models in the real-world has caused a surge of interest in explaining AI models' decisions to humans. Reinforcement Learning is not an exception in this regard. In this work, we propose a method for offering local explanations on risk in reinforcement learning. Our method only requires a log of previous interactions between the agent and the environment to create a state-transition model. It is designed to work on RL environments with either continuous or discrete state and action spaces. After creating the model, actions of any agent can be explained in terms of the features most influential in increasing or decreasing risk or any other desirable objective function in the locality of the agent. Through experiments, we demonstrate the effectiveness of the proposed method in providing such explanations.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.