The term “foundation model” was originally introduced by Bommasani et al. [57] from Stanford’s Human-Centered AI Institute (HAI). Foundation models are defined as “large-scale data-trained base models based on self-supervised or semi-supervised learning that can be adapted to multiple downstream tasks.” This paradigm shift enabled by foundation models holds significant importance, as it allows replacing multiple narrow task-specific models with broader and more general foundation models that require only a single training session for rapid adaptation to various applications. This approach not only facilitates rapid model development and delivers better performance for both in-domain and out-of-domain scenarios but also generates intelligence with emergent properties from large-scale foundation models.