CFM introduces efficient training objectives for CNFs, improving generative modeling tasks.
Continuous normalizing flows (CNFs) are a generative method for learning probability distributions from finite random samples. This work establishes non-asymptotic error bounds for the distribution estimator based on CNFs with linear interpolation and flow matching, under assumptions on the target distribution.
The paper proposes a novel method called Dual Flow Matching (DFM) for training Continuous Normalizing Flows (CNFs) more efficiently by avoiding computationally expensive interpolation steps used in previous methods while achieving superior performance in density estimation tasks.
본 논문에서는 연속 정규화 흐름(CNF) 학습 과정을 단순화하면서도 기존 방법의 제약을 극복하는 새로운 방법인 보간 없는 이중 흐름 매칭(DFM)을 제안합니다. DFM은 역방향 벡터 필드 모델을 추가적으로 활용하여 변환의 bijectivity를 보장하면서 별도의 보간 과정 없이 효율적인 학습을 가능하게 합니다.