Generative Adversarial
Generative Adversarial Networks (GANs) are a class of machine learning models designed to generate new data instances that resemble a training dataset. Current research focuses on improving GAN performance and stability across diverse applications, including image enhancement, speech synthesis, and data augmentation, often employing architectures like HiFi-GAN and variations of GANs combined with other neural network types (e.g., autoencoders, transformers). This work is significant due to GANs' ability to address data scarcity issues, improve the quality of synthetic data for various tasks, and enhance the robustness of AI systems against adversarial attacks.
Papers
BigVGAN: A Universal Neural Vocoder with Large-Scale Training
Sang-gil Lee, Wei Ping, Boris Ginsburg, Bryan Catanzaro, Sungroh Yoon
Benefits of Overparameterized Convolutional Residual Networks: Function Approximation under Smoothness Constraint
Hao Liu, Minshuo Chen, Siawpeng Er, Wenjing Liao, Tong Zhang, Tuo Zhao