Variational Autoencoding Discrete Diffusion with Enhanced Dimensional Correlations Modeling
By: Tianyu Xie , Shuchen Xue , Zijin Feng and more
Potential Business Impact:
Makes computers create better pictures and words.
Discrete diffusion models have recently shown great promise for modeling complex discrete data, with masked diffusion models (MDMs) offering a compelling trade-off between quality and generation speed. MDMs denoise by progressively unmasking multiple dimensions from an all-masked input, but their performance can degrade when using few denoising steps due to limited modeling of inter-dimensional dependencies. In this paper, we propose Variational Autoencoding Discrete Diffusion (VADD), a novel framework that enhances discrete diffusion with latent variable modeling to implicitly capture correlations among dimensions. By introducing an auxiliary recognition model, VADD enables stable training via variational lower bounds maximization and amortized inference over the training set. Our approach retains the efficiency of traditional MDMs while significantly improving sample quality, especially when the number of denoising steps is small. Empirical results on 2D toy data, pixel-level image generation, and text generation demonstrate that VADD consistently outperforms MDM baselines.
Similar Papers
Variational Masked Diffusion Models
Machine Learning (CS)
Makes computers better at filling in missing words.
Automated Learning of Semantic Embedding Representations for Diffusion Models
Machine Learning (CS)
Makes computers understand pictures better for learning.
Masked Diffusion Models are Secretly Learned-Order Autoregressive Models
Machine Learning (CS)
Teaches computers to create ordered text better.