AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Diffusion Fine-Tuning via Reparameterized Policy Gradient of the Soft Q-Function

Hyeongyu Kang, Jaewoo Lee, Woocheol Shin, Kiyoung Om, Jinkyoo Park

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
add consistency model as x0 predictor (ablation)SQDF · Aesthetic rewardaesthetic score: 7.10 -> 7.87Improved
add replay buffer (ablation)SQDF · Aesthetic rewardDreamSim diversity: 0.56 -> 0.58Improved
add replay buffer (ablation)SQDF · Aesthetic rewardaesthetic score: 8.06 -> 7.87Hurt
add consistency model (ablation)SQDF · Aesthetic rewardDreamSim diversity: 0.62 -> 0.58Hurt
use discount factor gamma<1 instead of gamma=1 (ablation)SQDF · Aesthetic rewardalignment and diversityImproved
use gamma<1 instead of gamma=1 (ablation)SQDF · Aesthetic rewardaesthetic scoreHurt
use 2-step DDIM instead of Tweedie one-step x0 prediction (ablation)SQDF · Aesthetic rewardoptimization speedImproved
use 4-step DDIM instead of 2-step (ablation)SQDF · Aesthetic rewardtraining stabilityHurt
use SQDF-UCB instead of SEIKO-UCBSQDF · Online black-box optimizationaesthetic score: 7.49 -> 7.87Improved
use SQDF-Bootstrap instead of SEIKO-BootstrapSQDF · Online black-box optimizationDreamSim diversity: 0.24 -> 0.53Improved
use SQDF instead of KL-augmented DDPO/DRaFTSQDF · Text-to-image finetuningreward vs alignment/diversity trade-offImproved
use SQDF instead of ReFL/DRaFTSQDF · Text-to-image finetuningalignment and diversityImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy