Research map / Image generation / Text-to-image diffusion
Reinforcement Learning With LLMs Interaction for Distributed Diffusion Model Services
Hongyang Du, Ruichen Zhang, Dusit Tao Niyato, Jiawen Kang, Zehui Xiong, Shuguang Robert Cui, Xuemin Shen, Dong In Kim
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 |
|---|---|---|---|
| replace standard DDPG actor with GDM-based actor (G-DDPG) | DDPG under RLLI · distributed GDM AIGC resource allocation | total QoE: up to 15% | Improved |
| raise bit error probability above threshold | distributed GDM framework · two-device tiger/dog prompts | image quality: BEP below 6% -> above 10% | Hurt |
| increase server denoising steps t0 on a different-prompt device | distributed GDM framework · Device 2 dog image | tiger features leak: t0 = 2 -> t0 = 4 | Hurt |
| increase device denoising steps t2 after shared steps | distributed GDM framework · Device 2 dog image | tiger features removed: t2 > 6 when t0 = 3 | Improved |
| share denoising steps among users, saving three steps | distributed GDM framework · smartphone energy example | energy consumption: 20% to 40% | Improved |
| increase total energy budget from 8 to 12 denoising steps | allocation schemes · QoE requirement | achievable QoE: 8 steps no scheme above 0.8; 12 steps several above 0.8 | Improved |
| tune server steps for user personality (high agreeableness vs high openness) | distributed GDM framework · t2 = 8 with different t0 | QoE: t0 = 2 best vs t0 = 4 best | Mixed |
| increase server denoising steps when device runs a different prompt | G-DDPG · Device 2 QoE | QoE: moderate t0 best | Mixed |
| increase t0 and t1 when device and server share the same prompt | G-DDPG · Device 1 QoE | QoE | Improved |
Closest papers by meaning
- Training Diffusion Models with Reinforcement Learning (ICLR 2024 · similarity 0.84)
- Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models (NeurIPS 2026 · similarity 0.84)
- Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents (NeurIPS 2026 · similarity 0.84)
- Delving into RL for Image Generation with CoT: A Study on DPO vs. GRPO (NeurIPS 2025 · similarity 0.83)
- Grounded Answers for Multi-agent Decision-making Problem through Generative World Model (NeurIPS 2024 · similarity 0.83)
- Preference Adaptive and Sequential Text-to-Image Generation (ICML 2025 · similarity 0.83)
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.
