AIMarkTechPostabout 1 hour ago
Sakana AI Trains Networks Without Backpropagation
8 min read
Sakana AI's Diffusing Blame paper introduces Error Diffusion, a local learning rule that trains Dale-compliant dual-stream networks without backpropagation or weight transport. The method reaches 96.7% on MNIST and 61.7% on CIFAR-10. Each weight update depends only on presynaptic activity, postsynaptic activation, and a third local signal.
Level
Hype check
Tap to vote and see what everyone thinks.
