Humans perceive the world using a combination of two, three or all five senses. These sensory modalities are equivalent to various data modalities in computing terminology, such as text, images, audio and videos. Multimodal learning is a multi-disciplinary approach that can handle heterogeneity in data sources to build computer agents with intelligent capabilities. It involves combining multiple modalities to solve complex AI tasks, such as image captioning, visual question answering, and sentiment analysis. This field requires processing different input modalities simultaneously, each with its own representation, such as pixels for images or characters for text. Multimodal learning models use specialized embeddings and fusion modules to create unified representations of the data. The approach has several practical applications, including generating realistic visuals from text prompts, recognizing emotions in audiovisual cues, and improving image captioning accuracy. However, building efficient multimodal learning models is still a challenge due to the complexity of processing multiple modalities simultaneously, with issues such as high training times, limited interpretability, and inadequate evaluation metrics.