IDG member Daniel Karavolos attended the Foundations of Digital Games at Cape Cod in August 2017 to present his latest research at the Institute of Digital Games. His presentation took place at the Procedural Content Generation workshop, and highlighted ways in which game balance could be predicted from a visual representation of a game level combined with weapon parameters. The paper with the underlying details is "Learning the Patterns of Balance in a Multi-Player Shooter Game" which focuses on first person shooter games and especially on team-based deathplay matches: the input to the computational model is the image of the game level, highlighting areas of high cover and low cover (chest-high walls), and the parameters of the weapon used by each team (every member of the team always uses one weapon for simplicity). The most effective computational model is a convolutional neural network, taking advantage of the recent deep learning advances to achieve accuracy far higher than that of simpler networks or random guessing. More information, and the paper itself, can be found here.