This paper aims to theoretically analyze the complexity of feature transformations encoded in DNNs with ReLU layers. We propose metrics to measure three types of complexities of transformations based on the information theory. We further discover and prove the strong correlation between the complexity and the disentanglement of transformations. Based on the proposed metrics, we analyze two typical phenomena of the change of the transformation complexity during the training process, and explore the ceiling of a DNN's complexity. The proposed metrics can also be used as a loss to learn a DNN with the minimum complexity, which also controls the over-fitting level of the DNN and influences adversarial robustness, adversarial transferability, and knowledge consistency. Comprehensive comparative studies have provided new perspectives to understand the DNN.

Abstract. The goal of the project is to show that audio can successfully be the primary element of interactive entertainment by delivering pure audio experiences that demonstrate both the creative potential and emotional power of an audio experience. We develop two proofs of concept with the technical foundation supported by prototypes. The core technology is a combination of a 3D game engine and an audio engine used to build sound environments. The interactions are based on 3D trackers and surround sound headphones.

No abstract

No abstract