Vision-Language Models (VLMs), such as CLIP, LLaVA, and GPT-4V, have revolutionized multimedia applications, such as captioning, visual question answering, cross-modal retrieval, and multi-modal agents, by enabling open-vocabulary search and generation. While significant strides have been made in developing powerful VLMs, the field remains in the early stages of rigorously evaluating and understanding their real-world vulnerabilities, both empirically and theoretically. In response to this important but under-explored gap, the workshop aims to provide a forum for researchers to share advanced in theories, algorithms and applications of trustworthy multi-modal learning, fostering diverse viewpoints on core principles and emerging techniques for developing trustworthy VLMs in the wild.

We welcome all papers that are related but not limited to the following topics:

A. Theory:

• Uncertainty quantification for VLMs.
• Theoretical bounds on VLMs under distribution shifts.
• Causal perspectives on spurious correlations and multi-modal shortcut learning.
• Robust optimization and generalization bounds for multi-modal learning objectives

B. Algorithms:

• Out-of-distribution (OOD) generalization, OOD detection and test-time adaptation
• Detecting and mitigating hallucination from VLMs
• Robust training and inference against adversarial attacks, prompt injection and imperfect inputs.
• Post-hoc and intrinsic methods for explaining VLM decisions.
• Techniques for auditing and debiasing vision-language datasets.
• Methods to ensure fairness in the context of VLMs.
• Privacy-preserving machine learning for VLMs.

C. Benchmark and Application:

• Trustworthy usages of VLMs for scientific domains, including program languages, climate science, healthcare, life sciences, physics, and cognitive science.
• Detecting AI-generated content and deepfakes.
• New benchmarks and evaluation protocols for real-world trustworthiness