Research map / Continual learning / Online and replay-based
Continual Personalization for Diffusion Models
Yu-Chien Liao, Jr-Jen Chen, Chi-Pin Huang, Ci-Siang Lin, Meng-Lin Wu, Yu-Chiang Frank Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| select concept neurons instead of fine-tuning randomly picked neurons of equal number (ablation) | CNS · Single-concept personalization | CLIP-T: 73.30 -> 76.95 | Improved |
| select concept neurons instead of random neurons (ablation) | CNS · Multi-concept personalization | CLIP-T: 72.92 -> 79.22 | Improved |
| add neuron regularization loss L_reg (ablation) | CNS · Single-concept personalization | CLIP-I: 73.06 -> 74.88 | Improved |
| add neuron regularization loss L_reg (ablation) | CNS · Multi-concept personalization | CLIP-T: 77.02 -> 79.22 | Improved |
| fine-tune concept neurons incrementally instead of storing and fusing per-concept LoRAs | CNS · Multi-concept personalization (fusion cost) | fusion time (s): 727 -> 0 (vs Mix-of-Show) | Improved |
| CNS vs Mix-of-Show on single-concept image alignment | CNS · Single-concept personalization | CLIP-I: 75.86 -> 74.88 | Hurt |
| exclude general neurons from base neurons when selecting concept neurons | CNS · neuron selection | neuron fraction updated: about 5% of cross-attention neurons, 0.13% of parameters | Improved |
Closest papers by meaning
- SCNS: Continual Personalization of Diffusion Models via Submodular Concept Neuron Selection (ICML 2026 · similarity 0.94)
- How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (NeurIPS 2024 · similarity 0.89)
- Mining your own secrets: Diffusion Classifier Scores for Continual Personalization of Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.88)
- ConceptGuard: Continual Personalized Text-to-Image Generation with Forgetting and Confusion Mitigation (CVPR 2025 · similarity 0.87)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · similarity 0.85)
- Dynamic Expansion Diffusion Learning for Lifelong Generative Modelling (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
