# Steering Training Trajectories in Diffusion Representation Learning

DevFeed: [Steering Training Trajectories in Diffusion Representation Learning](<https://devfeed.tech/articles/steering-training-trajectories-in-diffusion-representation-learning-124451.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.05319>)

Author: Rajat Rasal, Avinash Kori, Tian Xia, Ben Glocker

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [autoencoder](<https://devfeed.tech/tags/autoencoder.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [html](<https://devfeed.tech/tags/html.md>), [image](<https://devfeed.tech/tags/image.md>), [learning](<https://devfeed.tech/tags/learning.md>), [models](<https://devfeed.tech/tags/models.md>), [noise](<https://devfeed.tech/tags/noise.md>), [object](<https://devfeed.tech/tags/object.md>), [quality](<https://devfeed.tech/tags/quality.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [steering](<https://devfeed.tech/tags/steering.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [training](<https://devfeed.tech/tags/training.md>), [with](<https://devfeed.tech/tags/with.md>)

## AI overview

Across diffusion autoencoder training runs, similar final reconstruction quality can accompany substantially different representation disentanglement. The authors identify two trajectory patterns: one improves image fidelity before disentanglement, while the other improves both more gradually. They propose SteeringDRL, which combines gated residual U-Nets to modulate shortcut pathways with a curriculum controlling early noise-level exposure. The abstract reports competitive attribute disentanglement and reduced sensitivity to random seed in many settings. With object-centric learning, the approach also improves unsupervised segmentation on synthetic and real-world datasets.

## Source excerpt

arXiv:2607.05319v2 Announce Type: replace-cross Abstract: Across training runs of a diffusion autoencoder, we observe similar image reconstruction quality, but substantially different levels of disentanglement in the learned representations. In this work, we study how training dynamics contribute to this behaviour. By analysing training trajectories in a controlled setting - curves of reconstruction quality against disentanglement during training - we identify two recurring trajectory patterns, which we call learning regimes. These regimes differ in how reconstruction and disentanglement evolve, despite similar reconstruction at the end of training. In the reconstruction regime, models improve image fidelity before disentanglement, while in the disentanglement regime, models improve both reconstruction and disentanglement more gradually. We hypothesise that modulating shortcut pathways in the diffusion U-Net and controlling early noise-level exposure can shape trajectories. To that end, we introduce SteeringDRL, combining gated residual U-Nets with a noise curriculum to steer models towards better disentanglement. Across attribute disentanglement benchmarks, SteeringDRL has competitive disentanglement, with reduced seed sensitivity in many settings. Our method extends to spatial disentanglement with object-centric learning, improving unsupervised segmentation on synthetic and real-world datasets.