Steering Away from Memorization: Reachability-Constrained Reinforcement Learning for Text-to-Image Diffusion
cs.CV, cs.AI, cs.LG
Submitted: 2026-02-24
Updated: 2026-11-11
Project page: https://s-karnik.github.io/rads-memorization-project-page
Terminology
Sources
- Training Diffusion Models with Reinforcement Learning
- Plug and Play Language Models: A Simple Approach to Controlled Text Generation
- MemBench: Memorized Image Trigger Prompt Dataset for Diffusion Models
- Preemptive Detection and Steering of LLM Misalignment via Latent Reachability
- How Diffusion Models Memorize
- DyME: Dynamic Multi-Concept Erasure in Diffusion Models with Bi-Level Orthogonal LoRA Adaptation
- Steering Your Diffusion Policy with Latent Space Reinforcement Learning
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs