Diffusion Model Based Consistency Network for ML Solver

In recent years, diffusion models have emerged as one of the leading methods for generative modeling in various domains, such as image generation, speech generation, and super-resolution. These models follow a two-step process: a forward step, where noise is added in a Markovian manner, followed by a reverse denoising step learned using a deep neural network. In our work, we propose a diffusion model-based consistency network for our in-house ML solver, CoAEMLSim (Ranade et al., 2022). The proposed model is tested on a Poisson 3-D dataset and shows a significant speed-up compared to the classical method of fixed-point iteration. We aim to present results on additional use cases in the final presentation.