Explicit Flow Matching: On The Theory of Flow Matching Algorithms with Applications

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Ryzhakov, Gleb, Pavlova, Svetlana, Sevriugov, Egor, Oseledets, Ivan
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914855388184576
author Ryzhakov, Gleb
Pavlova, Svetlana
Sevriugov, Egor
Oseledets, Ivan
author_facet Ryzhakov, Gleb
Pavlova, Svetlana
Sevriugov, Egor
Oseledets, Ivan
contents This paper proposes a novel method, Explicit Flow Matching (ExFM), for training and analyzing flow-based generative models. ExFM leverages a theoretically grounded loss function, ExFM loss (a tractable form of Flow Matching (FM) loss), to demonstrably reduce variance during training, leading to faster convergence and more stable learning. Based on theoretical analysis of these formulas, we derived exact expressions for the vector field (and score in stochastic cases) for model examples (in particular, for separating multiple exponents), and in some simple cases, exact solutions for trajectories. In addition, we also investigated simple cases of diffusion generative models by adding a stochastic term and obtained an explicit form of the expression for score. While the paper emphasizes the theoretical underpinnings of ExFM, it also showcases its effectiveness through numerical experiments on various datasets, including high-dimensional ones. Compared to traditional FM methods, ExFM achieves superior performance in terms of both learning speed and final outcomes.
format Preprint
id arxiv_https___arxiv_org_abs_2402_03232
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Explicit Flow Matching: On The Theory of Flow Matching Algorithms with Applications
Ryzhakov, Gleb
Pavlova, Svetlana
Sevriugov, Egor
Oseledets, Ivan
Machine Learning
This paper proposes a novel method, Explicit Flow Matching (ExFM), for training and analyzing flow-based generative models. ExFM leverages a theoretically grounded loss function, ExFM loss (a tractable form of Flow Matching (FM) loss), to demonstrably reduce variance during training, leading to faster convergence and more stable learning. Based on theoretical analysis of these formulas, we derived exact expressions for the vector field (and score in stochastic cases) for model examples (in particular, for separating multiple exponents), and in some simple cases, exact solutions for trajectories. In addition, we also investigated simple cases of diffusion generative models by adding a stochastic term and obtained an explicit form of the expression for score. While the paper emphasizes the theoretical underpinnings of ExFM, it also showcases its effectiveness through numerical experiments on various datasets, including high-dimensional ones. Compared to traditional FM methods, ExFM achieves superior performance in terms of both learning speed and final outcomes.
title Explicit Flow Matching: On The Theory of Flow Matching Algorithms with Applications
topic Machine Learning
url https://arxiv.org/abs/2402.03232