# Latent Traversals in Generative Models as Potential Flows

Source: [Latent Traversals in Generative Models as Potential Flows](<https://www.sota2.com/research/paper/68fa6fe61e209131cea47f93>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T01:29:17.533Z

Authors: Yue Song, T. Anderson Keller, Nicu Sebe, Max Welling

Published: 2023-04-25T15:53:45.000Z

## Abstract

Despite the significant recent progress in deep generative models, the underlying structure of their latent spaces is still poorly understood, thereby making the task of performing semantically meaningful latent traversals an open research challenge. Most prior work has aimed to solve this challenge by modeling latent structures linearly, and finding corresponding linear directions which result in \`disentangled' generations. In this work, we instead propose to model latent structures with a learned dynamic potential landscape, thereby performing latent traversals as the flow of samples down the landscape's gradient. Inspired by physics, optimal transport, and neuroscience, these potential landscapes are learned as physically realistic partial differential equations, thereby allowing them to flexibly vary over both space and time. To achieve disentanglement, multiple potentials are learned simultaneously, and are constrained by a classifier to be distinct and semantically self-consistent. Experimentally, we demonstrate that our method achieves both more qualitatively and quantitatively disentangled trajectories than state-of-the-art baselines. Further, we demonstrate that our method can be integrated as a regularization term during training, thereby acting as an inductive bias towards the learning of structured representations, ultimately improving model likelihood on similarly structured data.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Equivariance Error Measurement on Issac3D](<https://www.sota2.com/research/sota/equivariance-error-measurement-on-issac3d.md>): Equivariance Error Measurement; Issac3D; Robot X-move Error: 6572.19.

- [Unsupervised Disentanglement on Falcor3D (Physical Factors)](<https://www.sota2.com/research/sota/unsupervised-disentanglement-on-falcor3d-physical-factors.md>): Unsupervised Disentanglement; Falcor3D; Lighting Intensity: 8312.97.

- [Representation Learning on MNIST](<https://www.sota2.com/research/sota/representation-learning-on-mnist.md>): Representation Learning; MNIST; Scaling Equivariance Error: 234.78.

- [Disentanglement on Shapes3D (10% train)](<https://www.sota2.com/research/sota/disentanglement-on-shapes3d-10-percent-train.md>): Disentanglement; Shapes3D (10% train); VP Score: 91.48.

- [Disentanglement on Shapes3D 1% (train)](<https://www.sota2.com/research/sota/disentanglement-on-shapes3d-1-percent-train.md>): Disentanglement; Shapes3D 1% (train); VP Score: 72.32.

- [Disentanglement on MNIST (10% train)](<https://www.sota2.com/research/sota/disentanglement-on-mnist-10-percent-train.md>): Disentanglement; MNIST (10% train); VP Score: 93.05.

- [Disentanglement on MNIST (1% train set)](<https://www.sota2.com/research/sota/disentanglement-on-mnist-1-percent-train-set.md>): Disentanglement; MNIST (1% train set); VP Score: 91.27.

- [Equivariance Error on MNIST](<https://www.sota2.com/research/sota/equivariance-error-on-mnist.md>): Equivariance Error; MNIST; Equivariance Error (Scaling + Rotation): 582.17.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2304.12944>)

[PDF](<https://arxiv.org/pdf/2304.12944.pdf>)
