The recent advancements in contrastive learning for visual representations and point cloud understanding have shown significant strides in efficiency and adaptability. Researchers are increasingly focusing on developing methods that not only enhance the learning process but also reduce computational burdens. For instance, novel approaches like EPContrast have been introduced to manage the computational demands associated with large-scale point cloud data, while simultaneously improving the quality of learned representations. Similarly, SigCLR proposes a logistic loss-based contrastive learning method that competes with established SSL objectives, emphasizing the importance of learnable bias and fixed temperature settings. In the realm of in-pixel processing, circuits are being designed to perform adaptive contrast enhancement directly within pixel arrays, offering substantial improvements in image quality and real-time adaptability. Additionally, the rethinking of positive pairs in contrastive learning, as demonstrated by Hydra, suggests that learning from arbitrary pairs can yield superior performance and prevent dimensional collapse, broadening the scope of contrastive learning applications. These developments collectively indicate a trend towards more flexible, efficient, and powerful contrastive learning techniques that can handle diverse data types and scenarios.