GFlowNets and variational inference
cs.LG, stat.ML
Submitted: 2022-10-02
Updated: 2026-08-31
Comments: ICLR 2023 final version; code: https://github.com/GFNOrg/GFN_vs_HVI
Code: https://github.com/GFNOrg/GFN_vs_HVI
License: http://creativecommons.org/licenses/by/4.0/
The gist: This paper builds bridges between two families of probabilistic algorithms: (hierarchical) variational inference (VI), which is typically used to model distributions over continuous spaces, and
Terminology
Abstract
This paper builds bridges between two families of probabilistic algorithms: (hierarchical) variational inference (VI), which is typically used to model distributions over continuous spaces, and generative flow networks (GFlowNets), which have been used for distributions over discrete structures such as graphs. We demonstrate that, in certain cases, VI algorithms are equivalent to special cases of GFlowNets in the sense of equality of expected gradients of their learning objectives. We then point out the differences between the two families and show how these differences emerge experimentally. Notably, GFlowNets, which borrow ideas from reinforcement learning, are more amenable than VI to off-policy training without the cost of high gradient variance induced by importance sampling. We argue that this property of GFlowNets can provide advantages for capturing diversity in multimodal target distributions.
Sources
- Relational inductive biases, deep learning, and graph networks
- GFlowNet Foundations
- A theory of continuous generative flow networks
- Learning GFlowNets from partial episodes for improved convergence and stability
- Expectation Propagation for approximate Bayesian inference
- Unifying Generative Models with GFlowNets and Beyond
- Variational f-divergence Minimization
- A Variational Perspective on Generative Flow Networks
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks