Key points are not available for this paper at this time.
Given the multifaceted nature of reality, phenomena can be interpreted not only through singular perspectives but also by bringing together various dimensions. Meaning often emerges from the convergence of diverse perspectives, contexts, and forms of representation. The construction of systems capable of analyzing this multilayered structure requires integrating heterogeneous types of information within a holistic, interactive framework. For this multilayered structure to be processable by artificial intelligence systems, the synthesis of heterogeneous information types from various sources in a holistic structure is mandated. In response to this requirement, multimodal learning is an approach that aims to develop more contextual and generalizable artificial intelligence systems by combining heterogeneous data from different modalities (e.g., text, images, audio, sensor data) within an integrated structure. Based on recent literature, this review examines the conceptual foundations of multimodal learning and its key technical challenges, including representation learning, alignment, fusion, translation, missing modality, and co-learning. This study systematically compares and classifies more than 50 of the most prominent review articles published between 2010 and 2025 in a comprehensive table, summarizing the challenges they address, their application areas, and practical contributions. Attention has been drawn to areas often neglected in the literature, such as co-learning and missing modality, as well as to other critical gaps persisting in the field. Furthermore, the paper presents multimodal applications in healthcare, robotics, autonomous driving, remote sensing, and security, along with common multimodal datasets. By bridging theoretical foundations and real-world applications, this study provides a comprehensive reference for the field of multimodal learning.
Ateş et al. (Thu,) studied this question.