An Object Matrix Input Format for a Deep AI in Angry Birds and the Like

摘要

This paper describes how to improve the performance of a deep-convolution-neural-network AI by a new input format called object matrix in the popular game Angry Birds and the like. This method makes the AI agent only focus on game objects which are necessary for playing the game. Experimental results are provided that show a statistically significant increase of the neural network's accuracy by this input format in predicting positions to drag the current bird.