EmoReward self-evolving reward calibration advances emotional video RL
EmoReward distills reliability from training traces to adaptively allocate weights for facial motion and chromatic cues in video generation.
Stories per day · 30 days
2 RSS
Relatedaudio-video2evaluation2multimodal2reinforcement-learning2benchmark1huggingface1prompt-engineering1rl1robotics1video-generation1vlm1zero-shot1
12 stories, newest first · one card per story
Stories tagged #reward-model
D&A Enables VLMs as Zero-Shot Robotic Reward Models
Decouples progress estimation into objective description and rubric-based assessment to reduce confirmation bias in zero-shot VLM evaluations for robotics.
AVC-Reward: Event-Centric Audio-Video Consistency Evaluation
AVC-Reward scores complex audio-video consistency at event level and aligns with human judgments better than prior metrics.
AVC-Reward evaluates audio-video consistency at the event level
Authors say the reward model aligns best with human judgments and can steer a joint generator via online reinforcement learning.
LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof Verification
LeanPRM uses Lean to auto-label 152K proof steps, lifting a Qwen3-4B verifier's step accuracy from 48.2 to 74.7.
SyncReward: Human-Aligned Synchronization Reward for Audio-Visual Generation
SyncReward adds an 80,000-rating dataset and reward model for audio-visual synchronization.
Post-Training Frontier Text-to-Image Models by Composing Preference and Rubric Rewards
Composing preference and rubric rewards lifts post-trained Flux2dev and Ideogram-4 on the Arena text-to-image leaderboard.
Using Context Is Not Enough: Test-Time Training for Personalized Reward Modeling
P-TTT personalizes reward models with test-time fast weights, beating in-context preference conditioning without backprop.
Diffusion Reward Models
DRM models rewards with a diffusion transformer, capturing multimodal preferences and improving RLHF decisions.
BiasReducer: Adaptive Bias Mitigation for Reward Models
BiasReducer edits reward-model heads to cut length, confidence, and sycophancy bias without retraining.
Aligning One-Step Generative Models with Reward-Weighted Transport Distillation
RWTD post-trains one-step generators, raising SANA Sprint GenEval from 0.73 to 0.80.
RewardVerse: Rubric-Guided Policy Optimization for Video Reward Modeling
RewardVerse uses dynamic rubrics and RGPO to stabilize video reward scores and reduce scalar drift.