SALD: Self-Referenced Advantage Learning for Diffusion Models
cs.CV
Submitted: 2026-10-01
Updated: 2026-10-07
License: http://creativecommons.org/licenses/by/4.0/
The gist: Recent work on language-model adaptation has shown that single models can obtain informative training signals by evaluating their behavior in demonstrationor feedback-augmented contexts, with the
Terminology
Abstract
Recent work on language-model adaptation has shown that single models can obtain informative training signals by evaluating their behavior in demonstrationor feedback-augmented contexts, with the help of a teacher network, which is driven by the student's learned parameters. Inspired by this internal-reference principle, we investigate how diffusion models can identify self-referenced training signals without external demonstrations or teacher networks. We introduce SALD, a self-referenced training framework that evaluates each image-caption pair at two noise levels using the same model. The easier, lower-noise path is evaluated without gradient tracking to provide a reference, while the harder, higher-noise path provides the training gradient. Rather than directly distilling the easy-path prediction, SALD uses the difference between two path errors to adapt the hardpath objective. The proposed Advantage-Guided Diffusion (AGD) converts this relative error into a differentiable sample-level weight. Temporal Advantage Memory (TAM) accumulates relative difficulty across training and adapts the future gap between the two noise levels. Spectral Advantage Decomposition (SAD) further compares the residual power spectra of the two paths and constructs a differentiable, frequency-derived latent-element weight. All components share a single set of model parameters, requiring neither an external teacher network nor additional trainable parameters during training or inference, and no modification to the inference procedure. Experiments across multiple architectures and datasets demonstrate consistent improvements in generation quality, while component-wise ablations quantify the contributions of the proposed components.
Sources
- PixArt-\Sigma: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation
- Spectrally-Guided Diffusion Noise Schedules
- Aligning Distributionally Robust Optimization with Practical Deep Learning Needs
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- Reinforcement Learning via Self-Distillation
- No Other Representation Component Is Needed: Diffusion Transformers Can Provide Representation Guidance by Themselves
- Shaping Inductive Bias in Diffusion Models through Frequency-Based Noise Control
- Adaptively Point-weighting Curriculum Learning
- Flow Matching for Generative Modeling
- Not All Noises Are Created Equally:Diffusion Noise Selection and Optimization
- UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild
- Learning Transferable Visual Models From Natural Language Supervision
- Learning Deep Representations of Fine-grained Visual Descriptions
- Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
- Self-Distillation Enables Continual Learning
- Score-Based Generative Modeling through Stochastic Differential Equations
- An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models
- SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers
- Temporal Difference Learning for Diffusion Models
- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models