Regime-Aware Router Learning for Multi-Domain Medical Image Generation with Diffusion Experts
Heterogeneous medical image data motivate domain-specialized diffusion experts, but expert specialization alone does not determine how their predictions should be combined during sampling.
ProofPaper ↗
Key points
- Learning a global router is particularly challenging when the server has limited, potentially unrepresentative images.
- We study two router-learning strategies for a fixed expert pool.
- We evaluate these distinct signals through routing analyses and paired medical image generation.
- These findings establish server-data availability and expert performance as distinct considerations in learning routers for multi-domain diffusion models.
Sources (1)
- [1]Regime-Aware Router Learning for Multi-Domain Medical Image Generation with Diffusion ExpertsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 02:47 AM
Heterogeneous medical image data motivate domain-specialized diffusion experts, but expert specialization alone does not determine how their predictions should be combined during sampling.
Learning a global router is particularly challenging when the server has limited, potentially unrepresentative images.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 11, 2026GuidanceNFT: NFT-Based Reward Fine-Tuning of Guided Diffusion Models
- Oct 11, 2026Uncovering the Generative Landscape of Diffusion Models: A Spectral-Guided Quality-Diversity Diffusion Framework
- Oct 11, 2026Self-Representation Alignment in Diffusion Models for Video Background Music Generation
- Oct 11, 2026Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation
- Oct 10, 2026DiffGA: Steering Frozen Diffusion Models for Materials Design via Evolutionary Search
- Oct 10, 2026LightVAE: Towards Compact and Efficient Video Autoencoders