An MIT team published a study in PNAS showing how learning machinery can form from simple local connection rules. The proposed mechanism, called a Self-Assembling Motif, allows networks to organize without every component being explicitly designed. The model achieved performance similar to standard backpropagation on pattern identification tasks. Researchers caution that this computational result does not prove biological brains use the same mechanism.
Read the full story at modelcurrent.com.
Photo: Randgruppe / Pixabay
