UDT: Reconciling U-Nets and Diffusion Transformers with Data-Adaptive Token Reduction
Diffusion Transformers (DiTs) have emerged as a core architecture in generative modeling due to their scalability and adaptability to multimodal tasks. DiTs comprise isotropic transformer blocks, and learn representations progressively across depth, where the denoising objective drives later layers to focus on fine-detail reconstruction. This results in degraded representation quality and an imbalanced encoder-decoder behavior. Prior approaches such as representation alignment (REPA) mitigate th
Record details
Published: 2 August 2026
Source: arXiv cs.LG
Category: Research
Topics: Safety & alignment
Retrieved: 4 August 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
The Boy Who Cried Wolf: Adversarial Misclassification of Safe Inputs as Unsafe in Multimodal Guardrails
arXiv red teaming query · 2 August 2026
Moving the Safety Barrier: Dynamic Routing Adaptive Alignment Against White-Box Attacks
arXiv red teaming query · 2 August 2026
No Single Neuron of Failure: Distributed Safety Alignment Against White-Box Attacks
arXiv red teaming query · 2 August 2026
Scoring Rules! Statistical and Strategic Alignment for Text Evaluation Metrics
arXiv cs.LG · 2 August 2026
Slot2Text: Object-Centric Visual Tokenization for Efficient and Spatially Traceable Surgical MLLMs
arXiv cs.LG · 2 August 2026
A Data-Centric Perspective on Tree Visualizations
arXiv cs.HC · 2 August 2026
How to cite this record
ethics.ai (2 August 2026), “UDT: Reconciling U-Nets and Diffusion Transformers with Data-Adaptive Token Reduction,” evidence record 16155, https://ethics.ai/record/16155 (originally published by arXiv cs.LG).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.