Video Games: Human Decision and Learning – A typical scenario involves playing a video game, but the player is only allowed to make decisions at the beginning of the game, a consequence which can not be directly analyzed by the game designer. Despite that, a game designer can implement some actions, or even execute them, at the beginning of a game, to achieve a desired result. In this paper, we give a theoretical analysis of how this would be encoded as action representations in video games with the objective of learning video games using knowledge of player behaviors. Based on the notion of video games playing at the beginning, our model is able to extract and visualize the games state, actions, and actions of the player. Experiments performed on real game data show that our model can efficiently learn video games, allowing for a fast learning and prediction of the state and actions of the player. Our method compares favorably to current state-of-the-art games based on action representations (such as Action Replay and Strategy Replay), as well as the most recent state-of-the-art game learning methods that use knowledge of gameplay behavior on the task at hand.
Deep neural networks have become a popular approach for machine learning and visual recognition applications. This makes it very difficult to optimize training with these models. The goal of this paper is to study the effect of modeling over training data using different deep models and learning techniques. We used a deep neural network (DNN) model and a stochastic gradient descent classifier to explore which models outperform and learn the best performance. We compared the performance of learning the model and the algorithm using simulated data in which we used a variety of datasets. Experimental results showed that the difference was substantial.
On-line learning of spatiotemporal patterns using an exact node-distance approach
Structural Similarities and Outlier Perturbations
Video Games: Human Decision and Learning
Visual Tracking via Deep Generative ModelsDeep neural networks have become a popular approach for machine learning and visual recognition applications. This makes it very difficult to optimize training with these models. The goal of this paper is to study the effect of modeling over training data using different deep models and learning techniques. We used a deep neural network (DNN) model and a stochastic gradient descent classifier to explore which models outperform and learn the best performance. We compared the performance of learning the model and the algorithm using simulated data in which we used a variety of datasets. Experimental results showed that the difference was substantial.