Abstract
The task of conditional generation is one of the most important applicationsof generative models, and numerous methods have been developed to date based onthe celebrated diffusion models, with the guidance-based classifier-free methodtaking the lead. However, the theory of the guidance-based method not onlyrequires the user to fine-tune the "guidance strength," but its target vectorfield does not necessarily correspond to the conditional distribution used intraining. In this paper, we develop the theory of conditional generation basedon Flow Matching, a current strong contender of diffusion methods. Motivated bythe interpretation of a probability path as a distribution on path space, weestablish a novel theory of flow-based generation of conditional distributionby employing the mathematical framework of generalized continuity equationinstead of the continuity equation in flow matching. This theory naturallyderives a method that aims to match the matrix field as opposed to the vectorfield. Our framework ensures the continuity of the generated conditionaldistribution through the existence of flow between conditional distributions.We will present our theory through experiments and mathematical results.