"Self-Supervised Learning Through Efference Copies. (arXiv:2210.09224v2 [cs.LG] UPDATED)" — A paper which shows that the commonly used self-supervised learning approach of transforming each training datapoint into a pair of views, using the knowledge of this pairing as a positive (i.e. non-contrastive) self-supervisory sign, and potentially opposing it to unrelated, (i.e. contrastive) negative examples is an incomplete implementation of a concept from neuroscience, the Efference Copy (EC).

Paper: http://arxiv.org/abs/2210.09224

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Efference copy (EC). A) Sensory…