AI Video Game Legacy Assets Classifier

Abstract

Our department collects vast chunks of assorted data from legacy servers left over after teams disband and/or studios close. In most cases the domain experts that created that data and knew how to use it moved on resulting in a loss of the knowledge needed to make sense of it. Our solution to this issue was to train an AI model that could help identify and classify this data and subsequently feed the output to our catalog application. Our demonstration would show how our model was trained, how it carries out predictions, and how well it performs on those predictions. To illustrate performance we will present confusion matrices, F1 scores, and precision-recall graphs. We will also outline future plans regarding model improvements and uses.

Details

Creators
Stefan Serbicki
Institutions
Electronic Arts.
Date
Keywords
machine learning; data dumps; legacy; classification; artificial intelligence; video games
Publication Type
paper
License
CC BY 4.0 International
Download
299393 bytes

View This Publication