SSL

#self-supervised-learning
Typically methods make assumptions about the underlying data for a better decision function. These include:

• Smoothness Assumption: samples close together in feature space are likely to be from the same class
• Cluster Assumption: samples in a cluster are likely to be from the same class
• Low Density Assumption: class boundaries are likely to be in areas of the feature space that have lower density than the cluster

References

• Cheplygina, V., De Bruijne, M. and Pluim, J.P., 2019. Not-so-supervised: a survey of semi-supervised, multi-instance, and transfer learning in medical image analysis. Medical image analysis, 54, pp.280-296.