Simplified and Generalized Masked Diffusion for Discrete Data
This work addresses inefficiencies in generative modeling of discrete data for researchers and practitioners, offering a more effective alternative to autoregressive models, though it is incremental in improving existing masked diffusion approaches.
The paper tackles the problem of complex and suboptimal formulations in masked diffusion models for discrete data by introducing a simplified and general framework, resulting in models that surpass prior diffusion language models on perplexity and outperform autoregressive models on image modeling with bits per dimension of 2.75 on CIFAR-10 and 3.40 on ImageNet 64x64.
Masked (or absorbing) diffusion is actively explored as an alternative to autoregressive models for generative modeling of discrete data. However, existing work in this area has been hindered by unnecessarily complex model formulations and unclear relationships between different perspectives, leading to suboptimal parameterization, training objectives, and ad hoc adjustments to counteract these issues. In this work, we aim to provide a simple and general framework that unlocks the full potential of masked diffusion models. We show that the continuous-time variational objective of masked diffusion models is a simple weighted integral of cross-entropy losses. Our framework also enables training generalized masked diffusion models with state-dependent masking schedules. When evaluated by perplexity, our models trained on OpenWebText surpass prior diffusion language models at GPT-2 scale and demonstrate superior performance on 4 out of 5 zero-shot language modeling tasks. Furthermore, our models vastly outperform previous discrete diffusion models on pixel-level image modeling, achieving 2.75 (CIFAR-10) and 3.40 (ImageNet 64x64) bits per dimension that are better than autoregressive models of similar sizes. Our code is available at https://github.com/google-deepmind/md4.