How artificial neural networks build their own learning machinery [R]

MIT coverage (13 October 2026): https://news.mit.edu/2026/how-artificial-neural-networks-build-their-own-learning-machinery-1013 Paper: “How Biological Synapses Self-Assemble Gradient Learning” (PNAS, 9 October): https://www.pnas.org/doi/10.1073/pnas.2606245123 According to the MIT report, randomly interconnected neuron-like units governed by shared plasticity rules can assemble learning motifs and learn pattern recognition and image classification. The reported comparison finds performance similar to the backpropagation-trained networks tested. The distinction is between designing a learning circuit directly and specifying local rules from which its learning machinery can organise itself. The stability of the motifs under different initial connectivity and extension beyond the demonstrated tasks are useful questions for evaluating the result alongside classification accuracy. The authors explicitly caution that self-organisation does not establish that the brain uses this mechanism. The motivation is biological plausibility, rather than evidence of a replacement for backpropagation in large-scale models.