Image de couverture de livre

Why Does RL Generalize Better Than SFT? A Data-Centric Perspective on VLM Post-Training

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Lu, Aojun, Feng, Tao, Yuan, Hangjie, Li, Wei, Sun, Yanan
Format:	Preprint
Publié:	2026
Sujets:	Computer Vision and Pattern Recognition Machine Learning
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective
par: Lu, Aojun, et autres
Publié: (2025)

Revisiting Neural Networks for Continual Learning: An Architectural Perspective
par: Lu, Aojun, et autres
Publié: (2024)

Adapt before Continual Learning
par: Lu, Aojun, et autres
Publié: (2025)

C-Flat++: Towards a More Efficient and Powerful Framework for Continual Learning
par: Li, Wei, et autres
Publié: (2025)

When Does RL Help Medical VLMs? Disentangling Vision, SFT, and RL Gains
par: Jeddi, Ahmadreza, et autres
Publié: (2026)

Make Continual Learning Stronger via C-Flat
par: Bian, Ang, et autres
Publié: (2024)

CogFlow: Bridging Perception and Reasoning through Knowledge Internalization for Visual Mathematical Problem Solving
par: Chen, Shuhang, et autres
Publié: (2026)

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
par: Chu, Tianzhe, et autres
Publié: (2025)

First SFT, Second RL, Third UPT: Continual Improving Multi-Modal LLM Reasoning via Unsupervised Post-Training
par: Wei, Lai, et autres
Publié: (2025)

SimpleAR: Pushing the Frontier of Autoregressive Visual Generation through Pretraining, SFT, and RL
par: Wang, Junke, et autres
Publié: (2025)

GTR: Guided Thought Reinforcement Prevents Thought Collapse in RL-based VLM Agent Training
par: Wei, Tong, et autres
Publié: (2025)

RL makes MLLMs see better than SFT
par: Song, Junha, et autres
Publié: (2025)

OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles
par: Deng, Yihe, et autres
Publié: (2025)

Can Better Text Semantics in Prompt Tuning Improve VLM Generalization?
par: Kuchibhotla, Hari Chandana, et autres
Publié: (2024)

Does VLM Classification Benefit from LLM Description Semantics?
par: Ma, Pingchuan, et autres
Publié: (2024)

DreamRelation: Relation-Centric Video Customization
par: Wei, Yujie, et autres
Publié: (2025)

2Xplat: Two Experts Are Better Than One Generalist
par: Jeong, Hwasik, et autres
Publié: (2026)

Inlier-Centric Post-Training Quantization for Object Detection Models
par: Kim, Minsu, et autres
Publié: (2026)

Does Your 3D Encoder Really Work? When Pretrain-SFT from 2D VLMs Meets 3D VLMs
par: Li, Haoyuan, et autres
Publié: (2025)

A Faster Path to Continual Learning
par: Li, Wei, et autres
Publié: (2026)

ID-Crafter: VLM-Grounded Online RL for Compositional Multi-Subject Video Generation
par: Pan, Panwang, et autres
Publié: (2025)

Metis-RISE: RL Incentivizes and SFT Enhances Multimodal Reasoning Model Learning
par: Qiu, Haibo, et autres
Publié: (2025)

Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy
par: Ma, Xiaoxiao, et autres
Publié: (2025)

Are Object-Centric Representations Better At Compositional Generalization?
par: Kapl, Ferdinand, et autres
Publié: (2026)

SAMora: Enhancing SAM through Hierarchical Self-Supervised Pre-Training for Medical Images
par: Chen, Shuhang, et autres
Publié: (2025)

Ego-InBetween: Generating Object State Transitions in Ego-Centric Videos
par: Ge, Mengmeng, et autres
Publié: (2026)

Beyond SFT-to-RL: Pre-alignment via Black-Box On-Policy Distillation for Multimodal RL
par: Wang, Sudong, et autres
Publié: (2026)

Q-VLM: Post-training Quantization for Large Vision-Language Models
par: Wang, Changyuan, et autres
Publié: (2024)

Unprejudiced Training Auxiliary Tasks Makes Primary Better: A Multi-Task Learning Perspective
par: Li, Yuanze, et autres
Publié: (2024)

Small-Large Collaboration: Training-efficient Concept Personalization for Large VLM using a Meta Personalized Small VLM
par: Yang, Sihan, et autres
Publié: (2025)

CharTide: Data-Centric Chart-to-Code Generation via Tri-Perspective Tuning and Inquiry-Driven Evolution
par: Zheng, Xiangxi, et autres
Publié: (2026)

Enhancing Video Transformers for Action Understanding with VLM-aided Training
par: Lu, Hui, et autres
Publié: (2024)

Bi-VLM: Pushing Ultra-Low Precision Post-Training Quantization Boundaries in Vision-Language Models
par: Wang, Xijun, et autres
Publié: (2025)

ReasonGen-R1: CoT for Autoregressive Image generation models through SFT and RL
par: Zhang, Yu, et autres
Publié: (2025)

ThinkRL-Edit: Thinking in Reinforcement Learning for Reasoning-Centric Image Editing
par: Li, Hengjia, et autres
Publié: (2026)

ZeroFlow: Overcoming Catastrophic Forgetting is Easier than You Think
par: Feng, Tao, et autres
Publié: (2025)

Continual GUI Agents
par: Liu, Ziwei, et autres
Publié: (2026)

Frequency Autoregressive Image Generation with Continuous Tokens
par: Yu, Hu, et autres
Publié: (2025)

FREE: Faster and Better Data-Free Meta-Learning
par: Wei, Yongxian, et autres
Publié: (2024)

REO-VLM: Transforming VLM to Meet Regression Challenges in Earth Observation
par: Xue, Xizhe, et autres
Publié: (2024)