The mechanisms behind the success of multi-view self-supervised studying (MVSSL) usually are not but absolutely understood. Contrastive MVSSL strategies have been studied although the lens of InfoNCE, a decrease certain of the Mutual Info (MI). Nevertheless, the relation between different MVSSL strategies and MI stays unclear. We think about a distinct decrease certain on the MI consisting of an entropy and a reconstruction time period (ER), and analyze the principle MVSSL households by way of its lens. Via this ER certain, we present that clustering-based strategies resembling DeepCluster and SwAV maximize the MI. We additionally re-interpret the mechanisms of distillation-based approaches resembling BYOL and DINO, exhibiting that they explicitly maximize the reconstruction time period and implicitly encourage a steady entropy, and we affirm this empirically. We present that changing the goals of widespread MVSSL strategies with this ER certain achieves aggressive efficiency, whereas making all these strategies steady when coaching with smaller batch sizes.