Hebbian architecture AI model [R]

In three linesHebbian architecture AI model without backpropagation or gradients. Trained on CIFAR-10 over 50 epochs with 100k neurons. Uses only 5-7% of total parameters. Emergent behaviors: accuracy dips followed by jumps exceeding previous best, and recovery after intentional damage to active neurons and pathways.

Summary generated by Claude — human-verified