Abstract: Models for detecting payment outliers from healthcare claims often rely on sparse, high-dimensional feature vector encodings of diagnosis codes. These encodings tend to lose inherent ...