The rise of the metaverse has made representation learning essential for creating immersive virtual environments by modeling and synthesizing information across multiple sensory modalities. With the exponential growth in the volume and variety of data in our multimedia-rich era, understanding how to represent complex information through deep learning models has become a pivotal area of research. In this presentation, I will first explore foundational works in visual (image) representation learning, which span diverse image domains, and highlight how these models capture meaningful patterns in visual data. Then, I will discuss recent advancements in multi-modal representation learning, particularly the integration of image and text data, and how this contributes to a more comprehensive understanding of the world.