Diffusion Models
PulseAugur coverage of Diffusion Models — every cluster mentioning Diffusion Models across labs, papers, and developer communities, ranked by signal.
- 2026-06-05 research_milestone A new paper explores predicting human preference for text-to-image generations before creation. source
23 day(s) with sentiment data
-
TinySR diffusion model achieves real-time image super-resolution with 83% parameter reduction
Researchers have developed TinySR, a novel diffusion model designed for real-world image super-resolution that achieves real-time performance with significantly reduced computational cost and model size. The model emplo…
-
Normalizing Flows Prove Capable for Continuous Control in RL
Researchers have demonstrated that normalizing flows (NFs) are capable models for continuous control tasks in reinforcement learning (RL). Contrary to the prevailing belief that NFs lack sufficient expressivity, this pa…
-
Insurers leverage generative AI for catastrophe modeling amid risk concerns
The insurance industry is exploring the use of generative AI, specifically diffusion models, to improve catastrophe modeling and assess climate-related risks. These models can generate numerous plausible weather scenari…
-
New SharpMoE framework enhances diffusion models with accurate routing
Researchers have developed SharpMoE, a new framework designed to improve the efficiency and performance of Mixture-of-Experts (MoE) diffusion models used in visual generation. The framework addresses a routing assignmen…
-
Diffusion model learning curves analyzed for manifold data
A new research paper explores the theoretical underpinnings of diffusion models, specifically focusing on their learning curves when dealing with data distributed on low-dimensional manifolds. The study derives expressi…
-
DanceDuo platform uses diffusion models for AI-choreographed dance generation
A new platform called DanceDuo has been introduced, utilizing diffusion models to create AI-choreographed dance sequences synchronized with various music genres. This system enables users to select music, humanoid model…
-
LISA method accelerates AI model training for visual generation · 3 sources tracked
Researchers have introduced LISA (Likelihood Score Alignment), a novel regularization method designed to enhance the efficiency and performance of visual-condition controllable generation models. LISA works by explicitl…
-
UniTeD unifies perception and planning in autonomous driving with diffusion models · 3 sources tracked
Researchers have developed UniTeD, a novel framework that unifies perception and planning in autonomous driving using diffusion models. Unlike previous methods that decouple these tasks, UniTeD enables bidirectional inf…
-
Paper Unifies Diffusion Models and Flow Matching via Wasserstein Geometry
This paper explores the underlying geometry of diffusion models and flow matching, revealing that both are governed by the quadratic Wasserstein distance on the space of probability measures. The research posits that di…
-
New framework optimizes diffusion model guidance for better consistency-coverage trade-offs · 3 sources tracked
Researchers have developed a new information-theoretic framework to optimize classifier-free guidance (CFG) schedules in diffusion models. This approach aims to balance the trade-off between condition consistency and di…
-
Research: Diffusion models struggle with compositional generation tasks
A new research paper, "Catastrophic Compositional Generation: Why Vanilla Diffusion Models Fail to Extrapolate," published on arXiv, argues that standard conditional diffusion models struggle with compositional generati…
-
Diffusion models robustly adapt to low-dimensional structure
Researchers have demonstrated that diffusion models robustly adapt to low-dimensional data structures, accelerating sampling processes. Their theoretical framework shows that a wide range of update coefficients can achi…
-
New TooBad framework enables stealthy backdoor attacks on diffusion models
Researchers have developed a new backdoor attack framework called TooBad, specifically designed for diffusion models. This framework significantly enhances the performance of backdoor attacks by employing a novel trigge…
-
New AI methods tackle 3D and 4D reconstruction from single images and video
Two new research papers introduce novel approaches to 3D and 4D reconstruction from visual data. PRISM offers a feed-forward method for single-image 3D reconstruction, decomposing the task into a geometric prior and a l…
-
New training method boosts diffusion model robustness against data contamination
Researchers have developed a new training method for diffusion models that enhances their robustness against data contamination. By replacing the standard Mean Squared Error (MSE) denoising loss with a transformation de…
-
New CACFM method accelerates diffusion model inference using reinforcement learning
Researchers have developed a new method called Curvature-Adaptive Consistency Flow Matching (CACFM) to accelerate the inference of diffusion models. This approach uses a reinforcement learning agent to dynamically optim…
-
New diffusion model sampling method promotes diverse outputs
Researchers have developed a new method for sampling from diffusion models that encourages diversity among the generated outputs. This approach, called Variance-Tilted Diffusion Models, uses a variance-weighted batch di…
-
New variance-tilted diffusion models enhance sample diversity
Researchers have developed a new method called variance-tilted diffusion models to improve the diversity of samples generated by diffusion models. This approach introduces a variance-weighted batch distribution that enc…
-
UK Motor Insurance Fraud Evolves with Generative AI
Generative AI is enabling a new wave of sophisticated motor insurance fraud in the UK, moving beyond manual efforts to automated fabrication of evidence. Fraudsters are using tools like diffusion models to create realis…
-
New AI Framework Enhances Domain Generalization with Adversarial Prompt Tuning
This paper introduces a novel framework called Progressive Adversarial Prompt Tuning (PAPT) designed to enhance single-domain generalization (SDG) in AI models. PAPT leverages pre-trained text-to-image foundation models…