A team from MIT demonstrated that neural networks can form useful learning circuits from random connections using local plasticity rules. The study, published in the Proceedings of the National Academy of Sciences, introduces a framework called Self-Assembling Learning. The resulting networks achieved performance similar to those trained with backpropagation on image classification tasks. This approach shifts the design focus from prewiring specific circuits to defining local interaction rules. The findings offer a new perspective on how biological brains might organize learning mechanisms without a central architect.
Read the full story at remio.ai.
Photo: geralt / Pixabay
