Trustworthy Multimodal Intelligence: Robust, Adaptive, and Uncertainty-Aware Learning

Multimodal artificial intelligence has shown strong potential in integrating heterogeneous information from text, vision, audio, sensors, and other data sources. However, real-world multimodal systems often operate under imperfect conditions, including missing or noisy modalities, distribution shifts, conflicting information, limited data quality, and uncertain predictions. These challenges can significantly affect model robustness, reliability, and generalization. Importantly, robustness, uncertainty, adaptability, and interpretability are often closely interconnected in such settings, and weaknesses in one aspect may directly affect the others. As multimodal AI is increasingly deployed in high-stakes and human-centered applications, there is a growing need to move beyond performance-oriented learning toward trustworthy multimodal intelligence that can jointly address these challenges, adapt to changing environments, and provide reliable and well-calibrated predictions under complex and dynamic conditions.This Research Topic aims to advance trustworthy multimodal intelligence by investigating how robustness, uncertainty awareness, adaptability, and interpretability can be jointly modeled rather than treated as isolated properties. Particular emphasis is placed on their interactions under realistic imperfect-data conditions, where multimodal systems may encounter missing or conflicting modalities, distribution shifts, out-of-distribution samples, and poorly calibrated predictions. We encourage methods that not only improve predictive performance but also provide reliable uncertainty estimates, calibrated confidence, robust multimodal fusion, and adaptive behavior under changing environments. The Topic seeks to bridge methodological advances with practical requirements for dependable multimodal AI, including reliability, interpretability, generalization, efficiency, and safety in real-world deployment.We welcome Original Research, Review, Mini Review, Perspective, and related article types on trustworthy multimodal intelligence. Relevant themes include:- joint modeling of robustness, uncertainty, adaptation, and interpretability;- reliability-aware multimodal representation and fusion;- uncertainty quantification and calibration;- missing, noisy, or conflicting modalities;- distribution shift and out-of-distribution detection;- calibration failure and selective prediction;and evaluation protocols and metrics beyond predictive accuracy.Work on multimodal foundation models and vision-language models is particularly encouraged when trustworthiness, uncertainty, robustness, or multimodal reliability is central to the contribution. Federated, distributed, and continual multimodal learning are also within scope when they explicitly address trustworthy and reliable multimodal intelligence. Topic editors: Hui Chen, Hengyu Liu Source: https://www.frontiersin.org/research-topics/85956/trustworthy-multimodal-intelligence-robust-adaptive-and-uncertainty-aware-learning