Contrastive Multiview Coding (CMC) is a method for learning unsupervised representations from multiple views of data. The approach is based on contrastive learning, which aims to maximize mutual information between different views of the same scene while minimizing differences between views of different scenes. The method is view-agnostic and can scale to any number of views. CMC outperforms cross-view prediction methods and achieves state-of-the-art results on image and video benchmarks. The method is evaluated on multiple datasets, including ImageNet and STL-10, and is shown to improve representation quality as the number of views increases. CMC is also applied to video tasks, where it uses image and optical flow as views. The method is compared to other self-supervised learning approaches and is shown to produce high-quality representations. The paper also discusses the relationship between mutual information and representation quality, showing that optimal views share relevant information but not too much. The results demonstrate that CMC is effective for learning representations from multiple views and outperforms predictive learning methods.Contrastive Multiview Coding (CMC) is a method for learning unsupervised representations from multiple views of data. The approach is based on contrastive learning, which aims to maximize mutual information between different views of the same scene while minimizing differences between views of different scenes. The method is view-agnostic and can scale to any number of views. CMC outperforms cross-view prediction methods and achieves state-of-the-art results on image and video benchmarks. The method is evaluated on multiple datasets, including ImageNet and STL-10, and is shown to improve representation quality as the number of views increases. CMC is also applied to video tasks, where it uses image and optical flow as views. The method is compared to other self-supervised learning approaches and is shown to produce high-quality representations. The paper also discusses the relationship between mutual information and representation quality, showing that optimal views share relevant information but not too much. The results demonstrate that CMC is effective for learning representations from multiple views and outperforms predictive learning methods.