VLM4RWD

Overview

The VLM4RWD workshop @ NeurIPS 2025 invite high-quality submissions that identify key challenges, introduce novel methodologies, and push the boundaries of efficient and reliable vision language models toward the next generation of autonomous systems. In addition to research papers, we also welcome demos and poster abstracts aligned with the workshop theme.
Accepted papers will be presented as posters during the poster sessions. Selected works will also be highlighted as contributed talks. Topics of interest include, but are not limited to:

Topics

• Data pipelines for efficient multimodal learning, from pretraining, to finetuning, and ultimately test-time adaptation
• Approaches for accelerating VLMs inference through the design of algorithms and systems
• Compression and distillation for VLMs deployment under constrained resources
• Sparse, modular, and retrieval-augmented VLM architectures
• Effective and efficient training for complex reasoning tasks in visual scenes
• Robust training and evaluation of robustness of VLMs
• Benchmarks for deployment-oriented evaluation of VLMs
• Mitigating hallucination and improving multimodal grounding
• Emerging frontiers: agentic VLMs for real-world integration

Submission Guidelines

• Formatting Instructions: Workshop papers should be max 8 pages in length (excluding refrences and appendix), following the NeurIPS 2025 main conference format. Supplementary materials and appendices are allowed but will be considered optional for reviewers.
• Submission link: Submissions will be managed via: OpenReview submission system.
• Reviews: The review process will be double-blind. All submissions must be anonymized and the leakage of any identification information is prohibited.
• What can be sumitted: Consistent with common workshop practices, we also welcome demo papers (2-4 pages) and extended abstracts (2 pages). we also welcome submissions of papers previously published; however, they will not be eligible for awards and will be limited to poster session.