ABC: Non-Markovian Diffusion Bridges for Continuous-Time Processes

A new paper on arXiv proposes ABC (Any-Subset Autoregressive Models via Non-Markovian Diffusion Bridges) for generating continuous-time, continuous-space stochastic processes conditioned on partial observations. Existing diffusion models struggle with capturing structural similarity between close states, unstable integration, and insensitivity to physical time elapsed. ABC models the process with a single SDE whose time variable and intermediate states track real time, offering provable advantages. The approach enables conditioning on arbitrary subsets of states, such as irregularly sampled timesteps or future observations.